WO2024120409A1 - Ai网络模型的确定方法、信息传输方法、装置和通信设备 - Google Patents

Ai网络模型的确定方法、信息传输方法、装置和通信设备 Download PDF

Info

Publication number
WO2024120409A1
WO2024120409A1 PCT/CN2023/136611 CN2023136611W WO2024120409A1 WO 2024120409 A1 WO2024120409 A1 WO 2024120409A1 CN 2023136611 W CN2023136611 W CN 2023136611W WO 2024120409 A1 WO2024120409 A1 WO 2024120409A1
Authority
WO
WIPO (PCT)
Prior art keywords
network
information
network model
model
terminal
Prior art date
Application number
PCT/CN2023/136611
Other languages
English (en)
French (fr)
Inventor
孙布勒
孙鹏
Original Assignee
维沃移动通信有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 维沃移动通信有限公司 filed Critical 维沃移动通信有限公司
Publication of WO2024120409A1 publication Critical patent/WO2024120409A1/zh

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/16Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks using machine learning or artificial intelligence

Definitions

  • the present application belongs to the field of communication technology, and specifically relates to a method for determining an AI network model, an information transmission method, an apparatus, and a communication device.
  • AI artificial intelligence
  • the embodiments of the present application provide a method for determining an AI network model, an information transmission method, an apparatus, and a communication device, which can obtain a super network with a high degree of freedom based on Neural Architecture Search (NAS).
  • NAS Neural Architecture Search
  • a method for transmitting information comprising:
  • the terminal receives first information from a network side device, wherein the first information includes relevant parameters of a super network model or information of a first AI network model generated based on the super network model, the super network model includes at least one level, each level has at least one network block, and when there are multiple network blocks in the same level, at least two network blocks of the multiple network blocks are connected in parallel;
  • the terminal determines a target AI network model according to the first information.
  • a device for determining an AI network model comprising:
  • a first receiving module configured to receive first information from a network-side device, wherein the first information includes relevant parameters of a super-network model or information of a first AI network model generated based on the super-network model, wherein the super-network model includes at least one level, each level having at least one network block, and when there are multiple network blocks in the same level, at least two of the multiple network blocks are connected in parallel;
  • a first determination module is used to determine a target AI network model according to the first information.
  • an information transmission method comprising:
  • the network side device sends first information to the terminal, wherein the first information includes relevant parameters of the super network model or information of the first AI network model generated based on the super network model, the super network model includes at least one level, each level has at least one network block, and when there are multiple network blocks in the same level, at least two of the multiple network blocks are connected in parallel.
  • an information transmission device comprising:
  • a first sending module is used to send first information to a terminal, wherein the first information includes relevant parameters of a super network model or information of a first AI network model generated based on the super network model, the super network model includes at least one level, each level has at least one network block, and when there are multiple network blocks in the same level, at least two of the multiple network blocks are connected in parallel.
  • a communication device which includes a processor and a memory, wherein the memory stores a program or instruction that can be run on the processor, and when the program or instruction is executed by the processor, the steps of the method described in the first aspect or the third aspect are implemented.
  • a terminal comprising a processor and a communication interface, wherein the communication interface is used to receive first information from a network side device, wherein the first information includes relevant parameters of a super network model or information of a first AI network model generated based on the super network model, the super network model includes at least one level, each level has at least one network block, and when there are multiple network blocks in the same level, at least two of the multiple network blocks are connected in parallel; the processor is used to determine the target AI network model based on the first information.
  • a network side device including a processor and a communication interface, wherein the communication interface is used to send first information to a terminal, wherein the first information includes relevant parameters of a super network model or information of a first AI network model generated based on the super network model, and the super network model includes at least one level, each level having at least one network block, and when there are multiple network blocks in the same level, at least two of the multiple network blocks are connected in parallel.
  • a communication system comprising: a terminal and a network side device, wherein the terminal can be used to execute the steps of the information transmission method as described in the first aspect, and the network side device can be used to execute the steps of the information processing method as described in the third aspect.
  • a readable storage medium on which a program or instruction is stored.
  • the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented, or the steps of the method described in the third aspect are implemented.
  • a chip comprising a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run a program or instruction to implement the method described in the first aspect, or to implement the method described in the third aspect.
  • a computer program/program product is provided, wherein the computer program/program product is stored in a storage medium, and the computer program/program product is executed by at least one processor to implement the steps of the information transmission method as described in the first aspect, or the computer program/program product is executed by at least one processor to implement the steps of the information transmission method as described in the first aspect.
  • a terminal receives first information from a network-side device, wherein the first information includes relevant parameters of a super network model or information of a first AI network model generated based on the super network model, the super network model includes at least one level, each level has at least one network block, and when there are multiple network blocks in the same level, at least two of the multiple network blocks are connected in parallel; the terminal determines a target AI network model based on the first information.
  • the target AI network model obtained by the terminal is determined on the basis of the super network model, which can reduce the amount of calculation required to obtain the target AI network model compared to the method of training the target AI network model based on training samples in the related art.
  • FIG1 is a schematic diagram of the structure of a wireless communication system to which an embodiment of the present application can be applied;
  • FIG2 is a schematic diagram of the performance gain when the channel state information (CSI) is predicted based on the network model and when the CSI is not predicted;
  • CSI channel state information
  • FIG3 is a flow chart of a method for determining an AI network model provided in an embodiment of the present application.
  • FIG4 is a schematic diagram of the structure of a hypernetwork model
  • FIG. 5 is a schematic diagram of weight coefficients of a hypernetwork model.
  • FIG6 is a flow chart of an information transmission method provided in an embodiment of the present application.
  • FIG7 is a schematic diagram of the structure of an AI network model determination device provided in an embodiment of the present application.
  • FIG8 is a schematic diagram of the structure of an information transmission device provided in an embodiment of the present application.
  • FIG9 is a schematic diagram of the structure of a communication device provided in an embodiment of the present application.
  • FIG. 10 is a schematic diagram of the hardware structure of a terminal provided in an embodiment of the present application.
  • FIG. 11 is a schematic diagram of the structure of a network-side device provided in an embodiment of the present application.
  • first, second, etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way are interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by “first” and “second” are generally of the same type, and the number of objects is not limited.
  • the first object can be one or more.
  • “and/or” in the specification and claims represents at least one of the connected objects, and the character “/" generally represents that the objects associated with each other are in an "or” relationship.
  • LTE Long Term Evolution
  • LTE-A Long Term Evolution-Advanced
  • CDMA Code Division Multiple Access
  • TDMA Time Division Multiple Access
  • FDMA Frequency Division Multiple Access
  • OFDMA Orthogonal Frequency Division Multiple Access
  • SC-FDMA Single-carrier Frequency Division Multiple Access
  • NR New Radio
  • 6G 6th Generation
  • FIG1 shows a block diagram of a wireless communication system applicable to an embodiment of the present application.
  • the wireless communication system includes a terminal 11 and a network side device 12.
  • the terminal 11 may be a mobile phone, a tablet computer (Tablet Personal Computer), a laptop computer (Laptop Computer) or a notebook computer, a personal digital assistant (Personal Digital Assistant, PDA), a handheld computer, a netbook, an ultra-mobile personal computer (ultra-mobile personal computer, UMPC), a mobile Internet device (Mobile Internet Device, MID), an augmented reality (augmented reality, AR)/virtual reality (virtual reality, VR) device , robots, wearable devices (Wearable Device), vehicle user equipment (VUE), pedestrian user equipment (PUE), smart home (home appliances with wireless communication functions, such as refrigerators, televisions, washing machines or furniture, etc.), game consoles, personal computers (personal computers, PCs), teller machines or self-service machines and other terminal side devices, wearable devices include: smart watches, smart bracelets, smart headphones,
  • the network side device 12 may include access network equipment or core network equipment, wherein the access network equipment may also be called wireless access network equipment, wireless access network (Radio Access Network, RAN), wireless access network function or wireless access network unit.
  • the access network equipment may include a base station, a wireless local area network (WLAN) access point or a WiFi node, etc.
  • WLAN wireless local area network
  • the base station may be called a node B, an evolved node B (eNB), an access point, a base transceiver station (BTS), a radio base station, a radio transceiver, a basic service set (BSS), an extended service set (ESS), a home node B, a home evolved node B, a transmission/reception point (TRP) or some other appropriate term in the field.
  • eNB evolved node B
  • BTS base transceiver station
  • ESS extended service set
  • home node B a home evolved node B
  • TRP transmission/reception point
  • AI models have been widely used in various fields. There are many ways to implement AI models, such as neural networks, decision trees, support vector machines, Bayesian classifiers, etc. This application uses neural networks as an example for illustration, but does not limit the specific type of AI models.
  • the parameters of the neural network are optimized through optimization algorithms. Optimization algorithms are a type of algorithm that can help us minimize or maximize the objective function (sometimes called loss function).
  • the objective function is often a mathematical combination of model parameters and data. For example, given data X and its corresponding label Y, we construct a neural network model f(.). With the model After the model is constructed, the predicted output f(x) can be obtained based on the input x, and the difference between the predicted value and the true value (f(x)-Y) can be calculated, which is the loss function. Our goal is to find appropriate weights and biases to minimize the value of the above loss function. The smaller the loss value, the closer our model is to the actual situation.
  • the common optimization algorithms are basically based on the Back Propagation (BP) algorithm.
  • BP Back Propagation
  • 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.
  • the input sample is transmitted from the input layer, processed by each hidden layer layer by layer, and then transmitted to the output layer. If the actual output of the output layer does not match the expected output, it will enter the back propagation stage of the error.
  • the back propagation of the error is to propagate the output error layer by layer through the hidden layer to the input layer in some form, and distribute the error to all units of each layer, so as to obtain the error signal of each layer unit, and this error signal is used as the basis for correcting the weights of each unit.
  • This process of adjusting the weights of each layer of the signal forward propagation and the error back propagation is repeated.
  • the process of continuous adjustment of weights is the learning and training process of the network. This process continues until the error of the network output is reduced to an acceptable level, or until the preset number of learning times is reached.
  • the selected AI algorithms and models vary depending on the type of solution.
  • the main way to improve 5G network performance with AI is to enhance or replace existing algorithms or processing modules with algorithms and models based on neural networks.
  • algorithms and models based on neural networks can achieve better performance than those based on deterministic algorithms.
  • Commonly used neural networks include deep neural networks, convolutional neural networks, and recurrent neural networks. With the help of existing AI tools, neural networks can be built, trained, and verified.
  • CSI prediction can be performed based on the AI network model, that is, historical CSI is input into the AI model, the AI network model analyzes the time domain variation characteristics of the channel, and outputs the future CSI.
  • the performance gain such as the Normalized Mean Squared Error (NMSE)
  • NMSE Normalized Mean Squared Error
  • the corresponding neural network needs to be run on the terminal.
  • the wireless environment changes, the business execution changes, etc.
  • the model used on the terminal side also needs to change, evolve, and update.
  • the network side device can send the super network model to the terminal.
  • the terminal moves, the wireless environment changes, the service execution changes, etc., the AI network model of the terminal does not match the current communication environment or service execution.
  • the terminal can update the final target AI network model based on the super network model.
  • the network side device can determine the first AI network model required by the terminal based on the super network model according to the terminal needs, and send the first AI network model to the terminal.
  • the super network model pre-trained by the network side device is used to generate the first AI network model.
  • Model generate the AI network model required by the terminal in new scenarios and new services.
  • the embodiment of the present application can reduce the amount of calculation and delay occupied by the process of training the AI network model required by the terminal.
  • An embodiment of the present application provides a method for determining an AI network model, wherein the execution subject is a terminal.
  • the method for determining an AI network model executed by the terminal may include the following steps:
  • Step 301 The terminal receives first information from a network side device, wherein the first information includes relevant parameters of a super network model or information of a first AI network model generated based on the super network model, the super network model includes at least one level, each level has at least one network block, and when there are multiple network blocks in the same level, at least two of the multiple network blocks are connected in parallel.
  • the first information includes relevant parameters of a super network model or information of a first AI network model generated based on the super network model
  • the super network model includes at least one level, each level has at least one network block, and when there are multiple network blocks in the same level, at least two of the multiple network blocks are connected in parallel.
  • Step 302 The terminal determines a target AI network model according to the first information.
  • the process of determining the AI network model based on the hypernetwork model can be a reference to the process of determining the AI network model based on the Neural Architecture Search (NAS) technology.
  • NAS Neural Architecture Search
  • AutoML Automated machine learning
  • AutoML attempts to automatically learn these important steps related to features, models, optimization, and evaluation, so that machine learning models can be applied without manual intervention.
  • AutoML can be seen as a system with very strong learning and generalization capabilities for given data and tasks.
  • AutoML can be seen as designing a series of advanced control systems to operate machine learning models, so that the models can automatically learn the appropriate parameters and configurations without manual intervention.
  • NAS is a subfield of AutoML. researchers have shifted their focus from manually designing network structures to automating network structure design, freeing researchers from the complicated work of network design and enabling them to achieve or even exceed manually designed network architectures in classification accuracy.
  • the most widely used NAS method is the differentiable method based on the hypernetwork model.
  • the hypernetwork model is shown in FIG4 .
  • the hypernetwork model includes multiple levels (such as layer 1 to layer 20), and the total number of levels can be set manually.
  • Each level of the hypernetwork model is composed of multiple neural network blocks (referred to as "network blocks" or blocks in the following embodiments of the present application) in parallel, and multiple neural network blocks at different levels can be the same or different.
  • Each neural network block is defined in advance (generally a verified, small neural network with excellent performance).
  • the above-mentioned parallel connection of network blocks in the same level can be understood as: the input of the network blocks in the same level is the same, and the output result of the level is the weighted sum of the outputs of all network blocks in the level.
  • the output of each layer is the weighted sum of the outputs of all neural network blocks in the layer, that is, the product of the output result of the nth neural network block in the ith layer and the weight coefficient of the nth neural network block in the ith layer to obtain the first value. Then, the first values of the 9 neural network blocks in the ith layer are summed to obtain The result is the output of the i-th level.
  • the weight coefficient ⁇ n (i) is the variable to be learned, and ⁇ n (i) represents the weight coefficient of the n-th network block in the i-th level.
  • the training data is input into the super network model, and then trained by supervised learning, semi-supervised learning, etc.
  • trainable parameters There are two types of trainable parameters in the training process, namely the training parameters of each neural network block of each level itself (a set of parameters, depending on the structure of the neural network block) and the weight of each neural network block of each level (a real number).
  • the training parameters of each neural network block of each level itself (a set of parameters, depending on the structure of the neural network block)
  • the weight of each neural network block of each level a real number.
  • the final neural network is generated according to the weights. For example, you can select a neural network block with the largest weight at each level, and connect the selected neural network blocks of all levels in series to obtain the final neural network. You can also select a single or multiple neural network blocks at each level to form the final neural network according to your needs and the distribution characteristics of the weights.
  • the AI network mode required by the terminal device is constantly updated. For example, as the terminal moves, the communication environment changes, the service changes, etc., the AI network model required by the terminal also changes accordingly.
  • the method for determining the AI network model provided in the embodiment of the present application, it is possible to train an AI network model that matches the communication environment and service of the terminal based on the actual needs of the terminal on the basis of the super network model.
  • NAS can be used to generate a neural network model in new scenarios and new services, and there is a very high degree of freedom when using NAS to select the final network model from the super network model.
  • the target AI network model may include an AI network model that is updated as the terminal moves, the communication environment changes, the service changes, etc., that is, the network model is used as the basis for updating the AI network model, or the target AI network model may include an initial AI network model, that is, the network model is used as the basis for determining the initial AI network model.
  • the target AI network model is mainly an updated AI network model for illustration, and no specific limitation is made here.
  • the network side device may send relevant parameters of the hypernetwork model to the terminal, and the terminal may generate the required AI network model based on the hypernetwork model. For example, the terminal selects a neural network block with the largest weight coefficient from each level of the hypernetwork model to form the target AI network model.
  • the super network model can represent a network model including I levels, and each level in the network model includes at least one network block, wherein I is an integer greater than 1. If a level in the super network model includes multiple network blocks, then at least two network blocks in the level are connected in parallel to each other.
  • the network blocks in the super network model may be pre-trained or defined network blocks, such as a predefined network block set, which includes at least two network blocks.
  • the network blocks in each level of the super network model may be network blocks selected from the network block set, and different levels in the super network model may select the same or different network blocks.
  • each level in the super network model includes all network blocks in the network block set, and the weight coefficients of the same network block at different levels may be different.
  • the terminal may update the super network model according to instructions from the network side device to obtain a target AI network model.
  • the network side device may determine a first AI network model based on a pre-trained super network model according to the environment, business, needs, and other information of the terminal, and issue the first AI network model to the terminal, wherein the first AI network model may be an AI network model that matches the environment, business, needs, and other information of the terminal.
  • the terminal may determine the first AI network model as the target AI network model, or the terminal may perform fine-tuning on the basis of the first AI network model to obtain the target AI network model.
  • the relevant parameters of the above hypernetwork model may include at least one of the following:
  • First indication information wherein the first indication information indicates a location or an identifier of a connection point between adjacent levels in the hypernetwork model.
  • the number of network blocks in each level of the above super network model can be the same or different.
  • the i-th level has Ni network blocks.
  • i can be an integer from 1 to I, where I is the number of levels included in the super network model.
  • the first parameter may include parameters of the network block other than the weight coefficient.
  • the first parameter may be a parameter set. For example, if the network block includes a neural network model, the first parameter includes the weight (multiplicative coefficient), bias (additive coefficient), activation function and other parameters of each neuron in the neural network model. Based on the first parameter, the corresponding network block can be restored.
  • the above weight coefficients can represent the weights of the output results of different network blocks in the same layer.
  • the output result of a certain layer is the sum of the products of the output results of each network block in the layer and the corresponding weight coefficients.
  • the network blocks of each level of the AI network model can be selected according to the weight coefficients of each network block, for example: the one with the largest weight coefficient or at least two with the top weight coefficients in each level of the AI network model are selected, and the final AI network model is constructed based on the network blocks selected in all levels.
  • a network block may include at least one layer of neural network.
  • the first indication information indicates the position or identification of the connection point between adjacent layers in the super network model, so that the adjacent layers in the super network model can be clearly divided.
  • the first indication information includes:
  • the first coefficient is obtained by multiplying the output result of the network block by the weight coefficient of the network block, and then the result obtained by summing the first coefficients of all network blocks in the same layer is used as the position of the connection point between adjacent layers in the super network model.
  • the adjacent layers in the super network model can be divided according to the value indicated by the first indication information.
  • the terminal and the network side device reach an agreement on the second information, wherein the second information includes at least one of the following:
  • a network block set wherein the network block set includes network blocks of all levels in the super network model
  • the terminal and the network side device reach an agreement on the second information by at least one of a protocol agreement, a network side device instruction, a terminal report, etc.
  • the identifier of each network block in the network block set may be the number of each network block in the network block set. For example, if the network block set includes 10 network blocks, the 10 network blocks may be numbered 1 to 10 in sequence.
  • the identifier of the network block can also be other types of identifiers, such as: the type and parameters of the neural network model adopted by the network block, the number of neural network layers it has, and any other identifier that can uniquely identify the network block in the network block set, which is not specifically limited here.
  • the network side device and the terminal can reach an agreement on the network block set, so that when the network side device sends down the super network model, it can send down the identifier of the network block corresponding to each level in the super network model.
  • the relevant parameters of the super network model may include that the network blocks of the first level include ⁇ neural network block 3 ⁇ , the network blocks of the second level include ⁇ neural network block 1, neural network block 3, neural network block 8 ⁇ , and the network blocks of the third level include ⁇ neural network block 4 ⁇ .
  • the terminal can process the collected data into information that conforms to the format of the input information of the hypernetwork model, so as to facilitate the use of the collected data to determine the target AI network model from the hypernetwork model, and/or facilitate the input of the collected data into the AI network model determined based on the hypernetwork model.
  • the terminal can be informed of the meaning of the output information of the hypernetwork model or the AI network model determined based on the hypernetwork model, thereby facilitating subsequent processing of the output information, such as determining the CSI prediction result based on the output information.
  • the number of levels included in the hypernetwork model may be determined based on at least one of terminal capabilities, terminal requirements, user settings, etc., and is not specifically limited herein.
  • the method before the terminal receives the first information from the network side device, the method further includes:
  • the terminal sends first request information to the network side device, wherein the first request information is used to request the first information.
  • the terminal when the terminal has a demand for a target AI network model, such as a new application scenario or service, the terminal can send a first request message to the network side device to request the first information. In this way, the terminal can update or retrain the AI network model according to the requested first information to obtain a target AI network model that matches the new application scenario or service.
  • a target AI network model such as a new application scenario or service
  • the method before the terminal determines the target AI network model according to the first information, the method further includes:
  • the terminal receives second indication information from the network side device, wherein the second indication information is used to instruct the terminal to determine the target AI network model based on the super network model.
  • the terminal under the instruction of the network side device, performs an operation of determining the target AI network model based on the super network model.
  • the terminal may also actively request the network side device to allow the terminal to determine the operation of the target AI network model based on the super network model, which is not specifically limited here.
  • the terminal determines the target AI network model according to the first information, including:
  • the terminal selects Ni network blocks from the i-th level of the super network model, where i is an integer less than or equal to I, I is the number of levels included in the super network model, and Ni is a positive integer;
  • the terminal determines the target AI network model based on the network blocks selected from all levels of the super network model.
  • the Ni corresponding to each level of the hypernetwork model may be the same or different.
  • Ni corresponding to each level of the target AI network model is equal to 1.
  • a network block may be selected from each level of the super network model, that is, each level of the target AI network model has only one network block.
  • the Ni network blocks include network blocks with weight coefficients greater than a preset value among all network blocks in the i-th level. For example, a network block with the largest weight coefficient or at least two network blocks with the highest weight coefficients can be selected from each level of the super network model.
  • the terminal may select at least one network block from each level of the super network model, and generate a target AI network model based on the network blocks selected from all levels.
  • the terminal can select at least one network block from each level of the super network model, connect the selected network blocks in series according to the levels, and then fine-tune them according to the sample data obtained by its own measurements to obtain the target AI network model.
  • the method further includes:
  • the terminal sends third information to the network side device, wherein the third information includes a solution for determining the target AI network model based on the super network model.
  • determining the target AI network model based on the super network model may be to select at least one network block from each level of the super network model, and to connect in series the network blocks selected from all levels of the super network model to obtain the target AI network model.
  • the scheme for determining the target AI network model based on the super network model may be a scheme for how to select network blocks from each level of the super network model, or may include which network blocks or blocks to select from each level of the super network model. For example: the number of network blocks selected from each level of the super network model, and/or the selected network blocks have large weight coefficients in the same level and/or are supported by the terminal capabilities and/or Or business related, etc.
  • the scheme for determining the target AI network model based on the super network model can be described based on the identification of the network block.
  • the network blocks selected at the first level include ⁇ neural network block 3 ⁇
  • the network blocks selected at the second level include ⁇ neural network block 1, neural network block 3, neural network block 8 ⁇
  • the network blocks selected at the third level include ⁇ neural network block 4 ⁇ , and so on.
  • the third information includes an identifier of a network block at each level in the target AI network model.
  • the network side device can clearly know the target AI network model actually used by the terminal.
  • the terminal sends the third information to the network side device, so that the network side device can learn from the third information that the terminal determines the target AI network model based on the super network model, and then learn the target AI network model of the terminal.
  • the terminal can send the third information to the network side device without the need to hide the model, or the terminal selects a network block with the largest weight coefficient from each level of the super network model by default.
  • the terminal can select at least two network blocks from at least one level of the super network model and not send the third information to the network side device.
  • the target AI network model finally determined by the terminal is unknown to the network side device, thereby achieving model hiding.
  • the user equipment (UE) side generates the neural network to be used in the end based on the super network model.
  • the general principle is to select a neural network block with the largest weight at each level, and connect the selected neural networks of all levels in series to obtain the final neural network.
  • model hiding cannot be achieved, that is, the model used by the UE is completely exposed to the network side.
  • the UE can select a single or multiple neural network blocks at each level to form the final neural network according to the distribution characteristics of the weights.
  • the first level selects the neural network block with the largest weight
  • the second level selects the three neural network blocks with the top three weights in parallel
  • the third level selects the neural network block with the second largest weight, until all levels of neural network blocks are selected.
  • the method further includes:
  • the terminal sends fourth information to the network side device, wherein the fourth information indicates network blocks supported and/or unsupported by the terminal, and the network blocks in the super network model are network blocks supported by the terminal.
  • the terminal may determine which network blocks it can support and/or which network blocks it cannot support based on its own hardware and/or software configuration.
  • the terminal sends the fourth information to the network side device, so that the network side device can select only the network blocks supported by the terminal to determine the super network model in the process of determining the super network model. In this way, the adaptation of the super network model and the target AI network model determined based on the super network model to the terminal capabilities can be improved.
  • the method further includes:
  • the terminal sends fifth information to the network side device, wherein the fifth information indicates the terminal needs The model size and/or complexity of the target AI network model, wherein the first AI network model is an AI network model that meets the requirements of the terminal.
  • the model size may be described in bits.
  • the fifth information may indicate the maximum number of bits of the target AI network model required by the terminal.
  • the complexity may be described using floating point numbers, such as bits per second.
  • the fifth information may indicate the maximum floating point number of the target AI network model required by the terminal.
  • the terminal sends the fifth information to the network side device, and the network side device determines the first AI network model that meets the terminal requirements based on the super network model, and sends the first AI network model to the terminal.
  • the terminal can use the first AI network model as the target AI network model, or fine-tune the first AI network model to obtain the target AI network model. In this way, the target AI network model obtained by the terminal can be lightweight.
  • the information interaction between the terminal and the network side device in the embodiment of the present application can be air interface information interaction.
  • the information in the interaction process can be carried in at least one of the following signaling or information:
  • PUCCH Physical Uplink Control Channel
  • PUSCH Physical Uplink Shared Channel
  • the information in the interaction process may be carried in at least one of the following signaling or information:
  • MAC CE Media Access Control Element
  • Radio Resource Control (RRC) messages
  • Non-Access Stratum (NAS) messages
  • DCI Downlink Control Information
  • SIB System Information Block
  • PDCCH Physical Downlink Control Channel
  • PDSCH Physical Downlink Shared Channel
  • the terminal receives first information from a network side device, wherein the first information includes Related parameters of the super network model or information of the first AI network model generated based on the super network model, the super network model includes at least one level, each level has at least one network block, when there are multiple network blocks in the same level, at least two network blocks of the multiple network blocks are connected in parallel; the terminal determines the target AI network model according to the first information.
  • the target AI network model obtained by the terminal is determined on the basis of the super network model, which can reduce the amount of calculation for obtaining the target AI network model compared to the method of training the target AI network model based on training samples in the related art.
  • the information processing method provided in the embodiment of the present application may be executed by a network-side device. As shown in FIG6 , the information transmission method may include the following steps:
  • Step 601 The network side device sends first information to the terminal, wherein the first information includes relevant parameters of a super network model or information of a first AI network model generated based on the super network model, the super network model includes at least one level, each level has at least one network block, and when there are multiple network blocks in the same level, at least two of the multiple network blocks are connected in parallel.
  • the first information includes relevant parameters of a super network model or information of a first AI network model generated based on the super network model
  • the super network model includes at least one level, each level has at least one network block, and when there are multiple network blocks in the same level, at least two of the multiple network blocks are connected in parallel.
  • the meanings and functions of the first information, super network model, related parameters of the super network model, first AI network model, level, and network block in the embodiments of the present application are the same as those of the first information, super network model, related parameters of the super network model, first AI network model, level, and network block in the method embodiment shown in Figure 3, and are not specifically limited here.
  • the information transmission method in this embodiment corresponds to the method embodiment shown in Figure 3, with the difference that the information transmission method in this embodiment is a network side device, while the executor of the method embodiment shown in Figure 3 is a terminal.
  • the explanation of the information transmission method in this embodiment can refer to the relevant instructions in the method embodiment shown in Figure 3, which will not be repeated here, and the terminal and the network side device are configured with each other, for example: the network side device determines the super network model and/or the first AI network model, and sends the first information to the terminal, so as to jointly enable the terminal to obtain the target AI network model determined based on the super network model, thereby reducing the amount of calculation for the terminal to determine the target AI network model.
  • the relevant parameters of the hypernetwork model include at least one of the following:
  • First indication information where the first indication information indicates a location or an identifier of a connection point between adjacent layers in the hypernetwork model.
  • the first indication information includes:
  • the network side device and the terminal reach an agreement on the second information, wherein the second information includes at least one of the following:
  • a network block set wherein the network block set includes network blocks of all levels in the super network model
  • the method before the network side device sends the first information to the terminal, the method further includes:
  • the network side device receives first request information from the terminal, wherein the first request information is used to request the first information.
  • the method further includes:
  • the network side device sends second indication information to the terminal, wherein the second indication information is used to instruct the terminal to determine a target AI network model based on the super network model.
  • the method further includes:
  • the network side device receives third information from the terminal, wherein the third information includes a scheme for determining a target AI network model based on the super network model, the target AI network model is an AI network model used by the terminal, and at least one layer in the target AI network model includes at least two network blocks.
  • the third information includes an identifier of a network block at each level in the target AI network model.
  • the method further includes:
  • the network side device receives fourth information from the terminal, wherein the fourth information indicates a network block supported and/or a network block not supported by the terminal;
  • the network side device determines the super network model according to the fourth information and a set of network blocks, wherein the network blocks in the super network model are network blocks supported by the terminal in the set of network blocks.
  • the method further includes:
  • the network-side device receives fifth information from the terminal, wherein the fifth information indicates a model size and/or complexity of a target AI network model required by the terminal;
  • the network-side device determines the first AI network model according to the fifth information and the super network model, wherein the first AI network model is an AI network model that meets the requirements of the terminal.
  • the information transmission method provided in the embodiment of the present application corresponds to the method for determining the AI network model in the implementation as described in Figure 3, and can jointly enable the terminal to obtain the target AI network model determined based on the super network model, thereby reducing the amount of calculation for the terminal to determine the target AI network model.
  • the method for determining an AI network model provided in the embodiment of the present application may be executed by a device for determining an AI network model.
  • the method for determining an AI network model performed by a device for determining an AI network model is taken as an example to illustrate the device for determining an AI network model provided in the embodiment of the present application.
  • an AI network model determination device provided in an embodiment of the present application may be a device in a terminal. As shown in FIG. 7 , the AI network model determination device 700 may include the following modules:
  • the first receiving module 701 is used to receive first information from a network side device, wherein the first information includes Related parameters of a super network model or information of a first AI network model generated based on the super network model, wherein the super network model includes at least one level, each level has at least one network block, and when there are multiple network blocks in the same level, at least two of the multiple network blocks are connected in parallel;
  • the first determination module 702 is used to determine the target AI network model according to the first information.
  • the relevant parameters of the hypernetwork model include at least one of the following:
  • First indication information wherein the first indication information indicates a location or an identifier of a connection point between adjacent levels in the hypernetwork model.
  • the first indication information includes:
  • the terminal and the network side device reach an agreement on the second information, wherein the second information includes at least one of the following:
  • a network block set wherein the network block set includes network blocks of all levels in the super network model
  • the AI network model determination device 700 further includes:
  • the second sending module is used to send first request information to the network side device, wherein the first request information is used to request the first information.
  • the AI network model determination device 700 further includes:
  • the second receiving module is used to receive second indication information from the network side device, wherein the second indication information is used to instruct the terminal to determine the target AI network model based on the super network model.
  • the first determining module 702 is specifically configured to:
  • Ni network blocks from the i-th level of the super network model, where i is an integer less than or equal to I, I is the number of levels included in the super network model, and Ni is a positive integer;
  • the target AI network model is determined based on network blocks selected from all levels of the super network model.
  • Ni corresponding to each level of the target AI network model is equal to 1.
  • the Ni network blocks include network blocks whose weight coefficients are greater than a preset value among all network blocks in the i-th level.
  • the AI network model determination device 700 further includes:
  • the third sending module is used to send third information to the network side device, wherein the third information includes a solution for determining the target AI network model based on the super network model.
  • the third information includes an identifier of a network block at each level in the target AI network model.
  • the AI network model determination device 700 further includes:
  • a fourth sending module is used to send fourth information to the network side device, wherein the fourth information indicates the network blocks supported and/or unsupported network blocks by the terminal, and the network blocks in the super network model are the network blocks supported by the terminal.
  • the AI network model determination device 700 further includes:
  • a fifth sending module is used to send fifth information to the network side device, wherein the fifth information indicates the model size and/or complexity of the target AI network model required by the terminal, and the first AI network model is an AI network model that meets the requirements of the terminal.
  • the determination device of the AI network 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 an electronic device, such as an integrated circuit or a chip.
  • the electronic device may be a terminal, or may be other devices other than a terminal.
  • the terminal may include but is 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 the embodiment of the present application.
  • the AI network model determination device 700 provided in the embodiment of the present application can implement each process implemented by the terminal in the method embodiment shown in Figure 3, and can achieve the same beneficial effects. To avoid repetition, it will not be described here.
  • the information transmission method provided in the embodiment of the present application can be executed by an information transmission device.
  • the information transmission device provided in the embodiment of the present application is described by taking the information transmission method executed by the information transmission device as an example.
  • An information transmission device provided in an embodiment of the present application may be a device in a network-side device. As shown in FIG8 , the information transmission device 800 may include the following modules:
  • the first sending module 801 is used to send first information to the terminal, wherein the first information includes relevant parameters of the super network model or information of the first AI network model generated based on the super network model, the super network model includes at least one level, each level has at least one network block, when there are multiple network blocks in the same level, at least two of the multiple network blocks are connected in parallel.
  • the relevant parameters of the hypernetwork model include at least one of the following:
  • First indication information where the first indication information indicates a location or an identifier of a connection point between adjacent layers in the hypernetwork model.
  • the first indication information includes:
  • the network side device and the terminal reach an agreement on the second information, wherein the second information includes at least one of the following:
  • a network block set wherein the network block set includes network blocks of all levels in the super network model
  • the information transmission device 800 further includes:
  • the third receiving module is used to receive first request information from the terminal, wherein the first request information is used to request the first information.
  • the information transmission device 800 further includes:
  • a sixth sending module is used to send second indication information to the terminal, wherein the second indication information is used to instruct the terminal to determine a target AI network model based on the super network model.
  • the information transmission device 800 further includes:
  • a fourth receiving module is used to receive third information from the terminal, wherein the third information includes a scheme for determining a target AI network model based on the super network model, the target AI network model is an AI network model used by the terminal, and at least one layer in the target AI network model includes at least two network blocks.
  • the third information includes an identifier of a network block at each level in the target AI network model.
  • the information transmission device 800 further includes:
  • a fifth receiving module configured to receive fourth information from the terminal, wherein the fourth information indicates network blocks supported and/or network blocks not supported by the terminal;
  • the second determination module is used to determine the super network model according to the fourth information and the network block set, wherein the network blocks in the super network model are the network blocks supported by the terminal in the network block set.
  • the information transmission device 800 further includes:
  • a sixth receiving module configured to receive fifth information from the terminal, wherein the fifth information indicates a model size and/or complexity of a target AI network model required by the terminal;
  • a third determination module is used to determine the first AI network model according to the fifth information and the super network model, wherein the first AI network model is an AI network model that meets the needs of the terminal.
  • the information transmission device 800 provided in the embodiment of the present application can implement each process implemented by the network side device in the method embodiment shown in Figure 6, and can achieve the same beneficial effects. To avoid repetition, it will not be described here.
  • the embodiment of the present application further provides a communication device 900, including a processor 901 and a memory 902, wherein the memory 902 stores a program or instruction that can be run on the processor 901.
  • the communication device 900 is a terminal
  • the program or instruction is executed by the processor 901 to implement the various steps of the method embodiment shown in FIG3 , and can achieve the same technical effect.
  • the communication device 900 is a network side device
  • the program or instruction is executed by the processor 901 to implement the various steps of the method embodiment shown in FIG6 , and can achieve the same technical effect.
  • I will not go into details here.
  • An embodiment of the present application also provides a terminal, including a processor and a communication interface, wherein the communication interface is used to receive first information from a network side device, wherein the first information includes relevant parameters of a super network model or information of a first AI network model generated based on the super network model, the super network model includes at least one level, each level has at least one network block, and when there are multiple network blocks in the same level, at least two of the multiple network blocks are connected in parallel; the processor is used to determine the target AI network model based on the first information.
  • FIG. 10 is a schematic diagram of the hardware structure of a terminal implementing an embodiment of the present application.
  • the terminal 1000 includes but is not limited to: a radio frequency unit 1001, a network module 1002, an audio output unit 1003, an input unit 1004, a sensor 1005, a display unit 1006, a user input unit 1007, an interface unit 1008, a memory 1009 and at least some of the components of a processor 1010.
  • the terminal 1000 may also include a power source (such as a battery) for supplying power to each component, and the power source may be logically connected to the processor 1010 through a power management system, so as to implement functions such as managing charging, discharging, and power consumption management through the power management system.
  • a power source such as a battery
  • the terminal structure shown in FIG10 does not constitute a limitation on the terminal, and the terminal may include more or fewer components than shown in the figure, or combine certain components, or arrange components differently, which will not be described in detail here.
  • the input unit 1004 may include a graphics processing unit (GPU) 10041 and a microphone 10042, and the graphics processor 10041 processes the image data of the static picture or video obtained by the image capture device (such as a camera) in the video capture mode or the image capture mode.
  • the display unit 1006 may include a display panel 10061, and the display panel 10061 may be configured in the form of a liquid crystal display, an organic light emitting diode, etc.
  • the user input unit 1007 includes a touch panel 10071 and at least one of other input devices 10072.
  • the touch panel 10071 is also called a touch screen.
  • the touch panel 10071 may include two parts: a touch detection device and a touch controller.
  • Other input devices 10072 may include, but are not limited to, a physical keyboard, function keys (such as a volume control key, a switch key, etc.), a trackball, a mouse, and a joystick, which will not be repeated here.
  • the RF unit 1001 can transmit the data to the processor 1010 for processing; in addition, the RF unit 1001 can send uplink data to the network side device.
  • the RF unit 1001 includes but is not limited to an antenna, an amplifier, a transceiver, a coupler, a low noise amplifier, a duplexer, etc.
  • the memory 1009 can be used to store software programs or instructions and various data.
  • the memory 1009 may mainly include a first storage area for storing programs or instructions and a second storage area for storing data, wherein the first storage area may store an operating system, an application program or instruction required for at least one function (such as a sound playback function, an image playback function, etc.), etc.
  • the memory 1009 may include a volatile memory or a non-volatile memory, or the memory 1009 may include both volatile and non-volatile memories.
  • the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or an electrically erasable programmable read-only memory (EEPROM). Or flash memory.
  • ROM read-only memory
  • PROM programmable read-only memory
  • EPROM erasable programmable read-only memory
  • EEPROM electrically erasable programmable read-only memory
  • EEPROM electrically erasable programmable read-only memory
  • Volatile memory can be random access memory (Random Access Memory, RAM), static random access memory (Static RAM, SRAM), dynamic random access memory (Dynamic RAM, DRAM), synchronous dynamic random access memory (Synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (Double Data Rate SDRAM, DDRSDRAM), enhanced synchronous dynamic random access memory (Enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (Synch link DRAM, SLDRAM) and direct memory bus random access memory (Direct Rambus RAM, DRRAM).
  • RAM Random Access Memory
  • SRAM static random access memory
  • DRAM dynamic random access memory
  • DRAM synchronous dynamic random access memory
  • SDRAM double data rate synchronous dynamic random access memory
  • Double Data Rate SDRAM Double Data Rate SDRAM
  • ESDRAM enhanced synchronous dynamic random access memory
  • Synch link DRAM, SLDRAM synchronous link dynamic random access memory
  • Direct Rambus RAM Direct Rambus RAM
  • the processor 1010 may include one or more processing units; optionally, the processor 1010 integrates an application processor and a modem processor, wherein the application processor mainly processes operations related to an operating system, a user interface, and application programs, and the modem processor mainly processes wireless communication signals, such as a baseband processor. It is understandable that the modem processor may not be integrated into the processor 1010.
  • the radio frequency unit 1001 is used to receive first information from the network side device, wherein the first information includes relevant parameters of the super network model or information of the first AI network model generated based on the super network model, the super network model includes at least one level, each level has at least one network block, and when there are multiple network blocks in the same level, at least two network blocks of the multiple network blocks are connected in parallel;
  • Processor 1010 is used to determine a target AI network model according to the first information.
  • the relevant parameters of the hypernetwork model include at least one of the following:
  • First indication information wherein the first indication information indicates a location or an identifier of a connection point between adjacent levels in the hypernetwork model.
  • the first indication information includes:
  • the terminal and the network side device reach an agreement on the second information, wherein the second information includes at least one of the following:
  • a network block set wherein the network block set includes network blocks of all levels in the super network model
  • the radio frequency unit 1001 is further used to send first request information to the network side device, wherein the first request information is used to request the first information.
  • processor 1010 performs the step of determining a target AI network model according to the first information:
  • the radio frequency unit 1001 is also used to receive second indication information from the network side device, wherein the second indication information is used to instruct the terminal to determine the target AI network model based on the super network model.
  • the determining of the target AI network model according to the first information performed by the processor 1010 includes:
  • Ni network blocks from the i-th level of the super network model, where i is an integer less than or equal to I, I is the number of levels included in the super network model, and Ni is a positive integer;
  • the target AI network model is determined based on network blocks selected from all levels of the super network model.
  • Ni corresponding to each level of the target AI network model is equal to 1.
  • the Ni network blocks include network blocks whose weight coefficients are greater than a preset value among all network blocks in the i-th level.
  • At least one level in the target AI network model includes at least two network blocks:
  • the radio frequency unit 1001 is also used to send third information to the network side device, wherein the third information includes a solution for determining the target AI network model based on the super network model.
  • the third information includes an identifier of a network block at each level in the target AI network model.
  • the radio frequency unit 1001 is also used to send fourth information to the network side device, wherein the fourth information indicates network blocks supported and/or unsupported network blocks by the terminal, and the network blocks in the super network model are network blocks supported by the terminal.
  • the radio frequency unit 1001 is also used to send fifth information to the network side device, wherein the fifth information indicates the model size and/or complexity of the target AI network model required by the terminal, and the first AI network model is an AI network model that meets the requirements of the terminal.
  • the terminal 1000 provided in the embodiment of the present application can implement the various processes performed by the determination device 700 of the AI network model shown in Figure 7, and can achieve the same beneficial effects. To avoid repetition, they will not be described here.
  • An embodiment of the present application also provides a network side device, including a processor and a communication interface, wherein the communication interface is used to send first information to a terminal, wherein the first information includes relevant parameters of a super network model or information of a first AI network model generated based on the super network model, and the super network model includes at least one level, each level has at least one network block, and when there are multiple network blocks in the same level, at least two of the multiple network blocks are connected in parallel.
  • the network side device 1100 can implement each process performed by the information transmission device 800 shown in Figure 8, and can achieve the same technical effect, which will not be repeated here.
  • the embodiment of the present application also provides a network side device.
  • the network side device 1100 includes: an antenna 1101, a radio frequency device 1102, a baseband device 1103, a processor 1104 and a memory 1105.
  • the antenna 1101 is connected to the radio frequency device 1102.
  • the radio frequency device 1102 receives information through the antenna 1101 and sends the received information to the baseband device 1103 for processing.
  • the baseband device 1103 processes the information to be sent and sends it to the radio frequency device 1102.
  • the radio frequency device 1102 processes the received information and sends it out through the antenna 1101.
  • the method executed by the network-side device in the above embodiment may be implemented in the baseband device 1103, which includes a baseband processor.
  • the baseband device 1103 may include, for example, at least one baseband board, on which a plurality of chips are arranged, as shown in FIG. As shown, one of the chips is, for example, a baseband processor, which is connected to the memory 1105 through a bus interface to call the program in the memory 1105 to execute the network device operations shown in the above method embodiment.
  • the network side device may also include a network interface 1106, which is, for example, a Common Public Radio Interface (CPRI).
  • CPRI Common Public Radio Interface
  • the network side device 1100 of the embodiment of the present application also includes: instructions or programs stored in the memory 1105 and executable on the processor 1104.
  • the processor 1104 calls the instructions or programs in the memory 1105 to execute the method executed by each module shown in Figure 8 and achieves the same technical effect. To avoid repetition, it will not be repeated here.
  • An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored.
  • a program or instruction is stored.
  • the various processes of the method embodiment shown in Figure 3 or Figure 6 are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.
  • the processor is the processor in the terminal described in the above embodiment.
  • the readable storage medium includes a computer readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.
  • An embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the various processes of the method embodiment shown in Figure 3 or Figure 6, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.
  • the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.
  • the embodiments of the present application further provide a computer program/program product, which is stored in a storage medium, and is executed by at least one processor to implement the various processes of the method embodiment shown in Figure 3 or Figure 6, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.
  • An embodiment of the present application also provides a communication system, including: a terminal and a network side device, wherein the terminal can be used to execute the steps of the method for determining the AI network model as shown in Figure 3, and the network side device can be used to execute the steps of the information transmission method as shown in Figure 6.
  • the above embodiment method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware, but in many cases the former is more convenient.
  • the technical solution of the present application, or the part that contributes to the relevant technology can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present application.

Landscapes

  • Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Signal Processing (AREA)
  • Medical Informatics (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本申请公开了一种AI网络模型的确定方法、信息传输方法、装置和通信设备,属于通信技术领域,本申请实施例的AI网络模型的确定方法包括:终端接收来自网络侧设备的第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联;所述终端根据所述第一信息确定目标AI网络模型。

Description

AI网络模型的确定方法、信息传输方法、装置和通信设备
相关申请的交叉引用
本申请主张在2022年12月7日在中国提交的中国专利申请No.202211567286.2的优先权,其全部内容通过引用包含于此。
技术领域
本申请属于通信技术领域,具体涉及一种AI网络模型的确定方法、信息传输方法、装置和通信设备。
背景技术
在相关技术中,对在通信网络中的收发端分别部署人工智能(Artificial Intelligence,AI)网络模型,以借助AI网络模型来进行模型推理进行了研究。
但是,将AI网络模型应用于无线通信***时,会面临多种场景以及多种业务。这样,需要对每一场景下的每一种业务分别训练AI网络模型,因此,AI网络模型的训练过程的计算量大。
发明内容
本申请实施例提供一种AI网络模型的确定方法、信息传输方法、装置和通信设备,能够基于神经网络架构搜索(Neural Architecture Search,NAS)获取具有高自由度的超网络,从而通过对该超网络的调整,便可以得到适用于各种场景以及各种业务的AI网络模型,能够降低训练AI网络模型的计算量。
第一方面,提供了一种信息传输方法,该方法包括:
终端接收来自网络侧设备的第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联;
所述终端根据所述第一信息确定目标AI网络模型。
第二方面,提供了一种AI网络模型的确定装置,该装置包括:
第一接收模块,用于接收来自网络侧设备的第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联;
第一确定模块,用于根据所述第一信息确定目标AI网络模型。
第三方面,提供了一种信息传输方法,包括:
网络侧设备向终端发送第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联。
第四方面,提供了一种信息传输装置,该装置包括:
第一发送模块,用于向终端发送第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联。
第五方面,提供了一种通信设备,该通信设备包括处理器和存储器,所述存储器存储可在所述处理器上运行的程序或指令,所述程序或指令被所述处理器执行时实现如第一方面或第三方面所述的方法的步骤。
第六方面,提供了一种终端,包括处理器及通信接口,其中,所述通信接口用于接收来自网络侧设备的第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联;所述处理器用于根据所述第一信息确定目标AI网络模型。
第七方面,提供了一种网络侧设备,包括处理器及通信接口,其中,所述通信接口用于向终端发送第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联。
第八方面,提供了一种通信***,包括:终端和网络侧设备,所述终端可用于执行如第一方面所述的信息传输方法的步骤,所述网络侧设备可用于执行如第三方面所述的信息处理方法的步骤。
第九方面,提供了一种可读存储介质,所述可读存储介质上存储程序或指令,所述程序或指令被处理器执行时实现如第一方面所述的方法的步骤,或者实现如第三方面所述的方法的步骤。
第十方面,提供了一种芯片,所述芯片包括处理器和通信接口,所述通信接口和所述处理器耦合,所述处理器用于运行程序或指令,实现如第一方面所述的方法,或实现如第三方面所述的方法。
第十一方面,提供了一种计算机程序/程序产品,所述计算机程序/程序产品被存储在存储介质中,所述计算机程序/程序产品被至少一个处理器执行以实现如第一方面所述的信息传输方法的步骤,或者所述计算机程序/程序产品被至少一个处理器执行以实现如第 三方面所述的信息处理方法的步骤。
在本申请实施例中,终端接收来自网络侧设备的第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联;所述终端根据所述第一信息确定目标AI网络模型。终端所获取的目标AI网络模型是在超网络模型的基础上确定的,相较于相关技术中,基于训练样本来训练目标AI网络模型的方式而言,能够降低得出目标AI网络模型的计算量。
附图说明
图1是本申请实施例能够应用的一种无线通信***的结构示意图;
图2是基于网络模型进行预测信道状态信息(Channel State Information,CSI)和未预测CSI时的性能增益示意图;
图3是本申请实施例提供的一种AI网络模型的确定方法的流程图;
图4是超网络模型的结构示意图;
图5是超网络模型的权重系数的示意图。
图6是本申请实施例提供的一种信息传输方法的流程图;
图7是本申请实施例提供的一种AI网络模型的确定装置的结构示意图;
图8是本申请实施例提供的一种信息传输装置的结构示意图;
图9是本申请实施例提供的一种通信设备的结构示意图;
图10是本申请实施例提供的一种终端的硬件结构示意图
图11是本申请实施例提供的一种网络侧设备的结构示意图。
具体实施方式
下面将结合本申请实施例中的附图,对本申请实施例中的技术方案进行清楚描述,显然,所描述的实施例是本申请一部分实施例,而不是全部的实施例。基于本申请中的实施例,本领域普通技术人员所获得的所有其他实施例,都属于本申请保护的范围。
本申请的说明书和权利要求书中的术语“第一”、“第二”等是用于区别类似的对象,而不用于描述特定的顺序或先后次序。应该理解这样使用的术语在适当情况下可以互换,以便本申请的实施例能够以除了在这里图示或描述的那些以外的顺序实施,且“第一”、“第二”所区别的对象通常为一类,并不限定对象的个数,例如第一对象可以是一个,也可以是多个。此外,说明书以及权利要求中“和/或”表示所连接对象的至少其中之一,字符“/”一般表示前后关联对象是一种“或”的关系。
值得指出的是,本申请实施例所描述的技术不限于长期演进型(Long Term Evolution,LTE)/LTE的演进(LTE-Advanced,LTE-A)***,还可用于其他无线通信***,诸如码 分多址(Code Division Multiple Access,CDMA)、时分多址(Time Division Multiple Access,TDMA)、频分多址(Frequency Division Multiple Access,FDMA)、正交频分多址(Orthogonal Frequency Division Multiple Access,OFDMA)、单载波频分多址(Single-carrier Frequency Division Multiple Access,SC-FDMA)和其他***。本申请实施例中的术语“***”和“网络”常被可互换地使用,所描述的技术既可用于以上提及的***和无线电技术,也可用于其他***和无线电技术。以下描述出于示例目的描述了新空口(New Radio,NR)***,并且在以下大部分描述中使用NR术语,但是这些技术也可应用于NR***应用以外的应用,如第6代(6th Generation,6G)通信***。
图1示出本申请实施例可应用的一种无线通信***的框图。无线通信***包括终端11和网络侧设备12。其中,终端11可以是手机、平板电脑(Tablet Personal Computer)、膝上型电脑(Laptop Computer)或称为笔记本电脑、个人数字助理(Personal Digital Assistant,PDA)、掌上电脑、上网本、超级移动个人计算机(ultra-mobile personal computer,UMPC)、移动上网装置(Mobile Internet Device,MID)、增强现实(augmented reality,AR)/虚拟现实(virtual reality,VR)设备、机器人、可穿戴式设备(Wearable Device)、车载设备(Vehicle User Equipment,VUE)、行人终端(Pedestrian User Equipment,PUE)、智能家居(具有无线通信功能的家居设备,如冰箱、电视、洗衣机或者家具等)、游戏机、个人计算机(personal computer,PC)、柜员机或者自助机等终端侧设备,可穿戴式设备包括:智能手表、智能手环、智能耳机、智能眼镜、智能首饰(智能手镯、智能手链、智能戒指、智能项链、智能脚镯、智能脚链等)、智能腕带、智能服装等。需要说明的是,在本申请实施例并不限定终端11的具体类型。网络侧设备12可以包括接入网设备或核心网设备,其中,接入网设备也可以称为无线接入网设备、无线接入网(Radio Access Network,RAN)、无线接入网功能或无线接入网单元。接入网设备可以包括基站、无线局域网(Wireless Local Area Network,WLAN)接入点或WiFi节点等,基站可被称为节点B、演进节点B(Evolved Node B,eNB)、接入点、基收发机站(Base Transceiver Station,BTS)、无线电基站、无线电收发机、基本服务集(Basic Service Set,BSS)、扩展服务集(Extended Service Set,ESS)、家用B节点、家用演进型B节点、发送接收点(Transmission/Reception Point,TRP)或所属领域中其他某个合适的术语,只要达到相同的技术效果,所述基站不限于特定技术词汇,需要说明的是,在本申请实施例中仅以NR***中的基站为例进行介绍,并不限定基站的具体类型。
人工智能目前在各个领域获得了广泛的应用。AI模型有多种实现方式,例如神经网络、决策树、支持向量机、贝叶斯分类器等。本申请以神经网络为例进行说明,但是并不限定AI模型的具体类型。
神经网络的参数通过优化算法进行优化。优化算法就是一种能够帮我们最小化或者最大化目标函数(有时候也叫损失函数)的一类算法。而目标函数往往是模型参数和数据的数学组合。例如给定数据X和其对应的标签Y,我们构建一个神经网络模型f(.),有了模 型后,根据输入x就可以得到预测输出f(x),并且可以计算出预测值和真实值之间的差距(f(x)-Y),这个就是损失函数。我们的目的是找到合适的权值和偏置,使上述的损失函数的值达到最小,损失值越小,则说明我们的模型越接近于真实情况。
目前常见的优化算法,基本都是基于误差反向传播(Back Propagation,BP)算法。BP算法的基本思想是,学习过程由信号的正向传播与误差的反向传播两个过程组成。正向传播时,输入样本从输入层传入,经各隐层逐层处理后,传向输出层。若输出层的实际输出与期望的输出不符,则转入误差的反向传播阶段。误差反传是将输出误差以某种形式通过隐层向输入层逐层反传,并将误差分摊给各层的所有单元,从而获得各层单元的误差信号,此误差信号即作为修正各单元权值的依据。这种信号正向传播与误差反向传播的各层权值调整过程,是周而复始地进行的。权值不断调整的过程,也就是网络的学习训练过程。此过程一直进行到网络输出的误差减少到可接受的程度,或进行到预先设定的学习次数为止。
一般而言,根据解决类型不同,选取的AI算法和采用的模型也有所差别。根据目前发表文章及公开研究成果,借助AI提升5G网络性能的主要方法是通过基于神经网络的算法和模型增强或者替代目前已有的算法或处理模块。在特定场景下,基于神经网络的算法和模型可以取得比基于确定性算法更好的性能。比较常用的神经网络包括深度神经网络、卷积神经网络和循环神经网络等。借助已有AI工具,可以实现神经网络的搭建、训练与验证工作。
通过AI或机器学习(Machine Learning,ML)方法替代相关技术的通信***中的模块能够有效提升***性能。例如:可以基于AI网络模型进行CSI预测,即将历史CSI输入给AI模型,AI网络模型分析信道的时域变化特性,并输出未来的CSI。如图2所示,在采用AI网络模型对未来不同时刻的CSI进行预测时,其取得的性能增益(如标准化均方误差(Normalized Mean Squared Error,NMSE))相较于不预测CSI的方案而言,有很大的提升,且预测的未来时刻不同,可以达到的预测精度也会不一样。
AI网络模型应用于无线通信***中时,需要在终端上运行相应的神经网络。但是,随着终端的移动,无线环境的变化、执行业务的变化等,终端侧使用的模型也需要进行变化、演进、更新。
在相关技术中,在重训练或更新AI网络模型时,需要使用大量的带标签的训练样本数据来训练最新的AI网络模型,并将训练好的AI网络模型同步至终端和网络侧设备,该过程中训练最新的AI网络模型的过程需要占用大量的计算资源且时延较长。
而本申请实施例中,网络侧设备可以将超过网络模型下发至终端,随着终端的移动、无线环境的变化、执行业务的变化等,使终端具有的AI网络模型与当前的通信环境或执行业务不匹配时,终端能够在超网络模型的基础上,更新得到最终的目标AI网络模型,或者,网络侧设备可以根据终端需求在超网络模型的基础上确定终端需要的第一AI网络模型,并将第一AI网络模型下发给终端。该过程中,基于网络侧设备预先训练的超网络 模型,在新场景和新业务中产生终端需要的AI网络模型。本申请实施例能够降低训练终端需要的AI网络模型的过程所占用的计算量和时延。
下面结合附图,通过一些实施例及其应用场景对本申请实施例提供的AI网络模型的确定方法、信息传输方法、AI网络模型的确定装置、信息传输装置及通信设备等进行详细地说明。
请参阅图3,本申请实施例提供的一种AI网络模型的确定方法,其执行主体是终端,如图3所示,该终端执行的AI网络模型的确定方法可以包括以下步骤:
步骤301、终端接收来自网络侧设备的第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联。
步骤302、所述终端根据所述第一信息确定目标AI网络模型。
值得提出的是,基于超网络模型来确定AI网络模型的过程,可以是参考基于神经网络架构搜索(Neural Architecture Search,NAS)技术来确定AI网络模型的过程。
其中,对于NAS的说明如下:
机器学习的应用需要大量的人工干预,这些人工干预表现在:特征提取、模型选择、参数调节等机器学习的各个方面。自动化机器学习(AutoML)试图将这些与特征、模型、优化、评价有关的重要步骤进行自动化地学习,使得机器学习模型无需人工干预即可被应用。从机器学习角度讲,AutoML可以看作是一个在给定数据和任务上学习和泛化能力非常强大的***。从自动化角度讲,AutoML则可以看作是设计一系列高级的控制***去操作机器学习模型,使得模型可以自动化地学习到合适的参数和配置而无需人工干预。
NAS是AutoML的一个子领域,研究者们将视线从手工设计网络结构转移到了网络结构设计自动化,使研究者们从繁复的网络设计工作上解脱出来,且能够在分类准确度上达到甚至超越手工设计的网络架构。
目前最广泛使用的NAS方法是基于超网络模型的可微分方法。
例如:超网络模型如图4所示,超网络模型包含多个层级(如:layer1~layer 20),总的层级数可以是人为设定的。超网络模型的每一层级都由多个神经网络块(本申请的以下实施例中称之为“网络块”或block)并联组成,不同层级的多个神经网络块可以相同也可以不同。每一个神经网络块是提前定义好的(一般是经过验证的,性能优异的小神经网络)。
上述同一层级内的网络块并联可以理解为:同一层级内的网络块的输入是同一个,且该层级的输出结果是本层级内的全部网络块的输出的加权和。
例如:如图5所示,每一个层级的输出是该层所有神经网络块的输出的加权和,即第i个层级的第n个神经网络块的输出结果与第i个层级的第n个神经网络块的权重系数的乘积,得到第一值,然后,将第i个层级内的9个神经网络块的第一值进行求和,得到的 结果便是第i个层级的输出。其中,权重系数αn (i)是待学习的变量,其中,αn (i)表示第i个层级中的第n个网络块的权重系数。将训练数据输入至超网络模型,再用监督学习、半监督学习等方式进行训练。训练的过程中可训练的参数有两种,即每个层级的每个神经网络块本身的训练参数(是一个参数集合,取决于该神经网络块的构造)和每个层级的每个神经网络块的权重(是一个实数)。将超网络模型训练好后,根据权重生成最终的神经网络。比如,可以在每一层级选取权重最大的一个神经网络块,将选出来的所有层级的神经网络快串联起来获得最终的神经网络。也可以按照自己的需求以及权重的分布特征在每一层级选取单个或多个神经网络块组成最终的神经网络。
值得提出的是,终端设备需要的AI网络模式是不断更新的,例如:随着终端的移动、所处通信环境的变化、业务变更等,终端需要的AI网络模型也随之发生变化,通过本申请实施例提供的AI网络模型的确定方法,可以在超网络模型的基础上,根据终端实际需要训练得到与终端所处通信环境的、业务等相匹配的AI网络模型,这样,可以使用NAS,在新场景和新业务中产生神经网络模型,且使用NAS从超网络模型选出最终的网络模型时,有非常高的自由度。
一种实施方式中,目标AI网络模型可以包括随着终端的移动、所处通信环境的变化、业务变更等而更新后的AI网络模型,即超过网络模型作为AI网络模型更新的基础,或者,目标AI网络模型可以包括初始的AI网络模型,即超过网络模型作为确定初始AI网络模型的基础。本申请实施例中,主要以目标AI网络模型是更新后的AI网络模型为了进行举例说明,在此不作具体限定。
一种实施方式中,网络侧设备可以向终端下发超网络模型的超网络模型的相关参数,终端则可以根据该超网络模型生成需要的AI网络模型。例如:终端从超网络模型的每一个层级中选择权重系数最大一个神经网络块,以构成目标AI网络模型。
其中,超网络模型可以表示包括I个层级的网络模型,且该网络模型中的每一个层级包括至少一个网络块,其中,I为大于1的整数,若超网络模型中的某一层级包括多个网络块,则该层级中的至少两个网络块是相互并联的。
一种实施方式中,超网络模型中的网络块可以是提前训练或定义好的网络块,如预定义网络块集合,该网络块集合包括至少两个网络块,超网络模型中的每一个层级中的网络块可以是从该网络块集合中选择的网络块,且超网络模型中的不同层级可以选择相同或者不同的网络块,例如:超网络模型中的每一个层级均包括网络块集合中的全部网络块,且位于不同层级的同一网络块的权重系数可以不同。
一种实施方式中,终端可以根据网络侧设备的指示,在超网络模型的基础上进行更新,得到目标AI网络模型。
一种实施方式中,网络侧设备可以根据终端所处的环境、业务、需求等信息,在预先训练的超网络模型的基础上,确定第一AI网络模型,并向终端下发该第一AI网络模型,其中,第一AI网络模型可以是与终端所处的环境、业务、需求等信息相匹配的AI网络模 型。此时,终端可以将第一AI网络模型确定为目标AI网络模型,或者,终端可以在第一AI网络模型的基础上进行微调,得到目标AI网络模型。
可选地,上述超网络模型的相关参数可以包括以下至少一项:
所述超网络模型中每一个层级具有的网络块的数目;
所述超网络模型中的每一个层级中的每一个网络块的第一参数;
所述超网络模型中的每一个层级中的每一个网络块的权重系数;
第一指示信息,所述第一指示信息指示所述超网络模型中的相邻层级之间的连接点的位置或标识。
选项一,上述超网络模型中每一个层级具有的网络块的数目,可以是相同的或不同的,例如:第i个层级具有Ni个网络块,此时,i可以是1至I的整数,I为所述超网络模型包括的层级的数量。
选项二,上述第一参数可以包括网络块的除了权重系数之外的参数,该第一参数可以是参数集合,如假设网络块包括神经网络模型,则第一参数包括该神经网络模型中的每一个神经元的权值(乘性系数)、偏置(加性系数)、激活函数等参数。基于该第一参数,可以还原对应的网络块。
选项三,上述权重系数可以表示同一层级中的不同网络块的输出结果所占的权重,例如:某一层级的输出结果为该层级中的每一个网络块的输出结果与对应的权重系数的乘积之和。
一种可选的实施方式中,在基于超网络模型来确定AI网络模型的过程中,可以根据各个网络块的权重系数来选择AI网络模型的每一个层级的网络块,例如:选择AI网络模型的每一个层级中权重系数最大的一个或权重系数靠前的至少两个,并根据全部层级中选择的网络块来构成最终的AI网络模型。
选项四,一种实施方式中,一个网络块可以包含至少一层神经网络,此时,通过第一指示信息指示所述超网络模型中的相邻层级之间的连接点的位置或标识,可以明确划分超网络模型中的相邻层级。
可选地,所述第一指示信息,包括:
所述超网络模型中的第一目标层级内的每一个网络块的输出信息与对应的权重系数的加权求和值,其中,所述超网络模型中的层包括所述第一目标层级。
本实施方式中,通过将网络块的输出结果与该网络块的权重系数相乘得到第一系数,然后,将同一层级内的全部网络块的第一系数进行求和得到的结果作为超网络模型中的相邻层级之间的连接点的位置。这样,在终端获知了超网络模型的相关参数的情况下,可以根据第一指示信息所指示的取值来划分超网络模型中的相邻层级。
作为一种可选的实施方式,所述终端与所述网络侧设备对第二信息达成一致,其中,所述第二信息包括以下至少一项:
所述超网络模型包含的层级的数量;
所述超网络模型的输入信息的格式;
所述超网络模型的输出信息的格式;
网络块集合,所述网络块集合包括所述超网络模型中的全部层级的网络块;
所述网络块集合中的每一个网络块的标识。
一种实施方式中,所述终端与所述网络侧设备对第二信息达成一致,可以是通过协议约定、网络侧设备指示、终端上报等方式中的至少一项来使所述终端与所述网络侧设备对第二信息达成一致。
一种实施方式中,所述网络块集合中的每一个网络块的标识,可以是所述网络块集合中的每一个网络块的编号,例如:网络块集合包括10个网络块,则可以依次对10个网络块编号为1至10。
当然,在实施中,网络块的标识还可以是其他类型的标识,如:网络块采用的神经网络模型的类型、参数、具有的神经网络的层数等任意能够唯一标识网络块集合中的网络块的标识,在此不作具体限定。
一种实施方式中,网络侧设备和终端可以实现对网络块集合达成一致,这样,在网络侧设备下发超网络模型时,可以下发超网络模型中的每一个层级对应的网络块的标识即可。例如:假设超网络模型包括3个层级,则超网络模型的相关参数可以包括第一层级的网络块包括{神经网络块3},第二层级的网络块包括{神经网络块1,神经网络块3,神经网络块8},第三层级的网络块包括{神经网络块4}。
一种实施方式中,通过约定所述超网络模型的输入信息的格式,可以使终端据此将采集数据处理成符合所述超网络模型的输入信息的格式的信息,以便于采用采集数据来从超网络模型中确定目标AI网络模型,和/或,便于将采集数据输入至基于超网络模型确定的AI网络模型中。
一种实施方式中,通过约定所述超网络模型的输出信息的格式,可以使终端获知所述超网络模型或基于超网络模型确定的AI网络模型的输出信息的含义,从而便于对该输出信息进行后续处理,如:基于输出信息确定CSI预测结果等。
一种实施方式中,超网络模型包含的层级的数量可以是根据终端能力、终端需求、用户设置情况等至少一项确定的,在此不作具体限定。
作为一种可选的实施方式,在所述终端接收来自网络侧设备的第一信息之前,所述方法还包括:
所述终端向所述网络侧设备发送第一请求信息,其中,所述第一请求信息用于请求所述第一信息。
本实施方式中,终端可以在具有目标AI网络模型的需求时,如终端具有的新的应用场景或业务时,向网络侧设备发送第一请求信息,以请求所述第一信息。这样,终端可以根据请求到的第一信息对AI网络模型进行更新或重训练,得到与新的应用场景或业务匹配的目标AI网络模型。
作为一种可选的实施方式,在所述终端根据所述第一信息确定目标AI网络模型之前,所述方法还包括:
所述终端接收来自所述网络侧设备的第二指示信息,其中,所述第二指示信息用于指示所述终端基于所述超网络模型来确定所述目标AI网络模型。
本实施方式中,终端在网络侧设备的指示下,执行基于所述超网络模型来确定所述目标AI网络模型的操作。
当然,在另一种实施方式中,终端也可以主动请求网络侧设备允许终端基于所述超网络模型来确定所述目标AI网络模型的操作,在此不作具体限定。
作为一种可选的实施方式,所述终端根据所述第一信息确定目标AI网络模型,包括:
所述终端从所述超网络模型的第i个层级中选择Ni个网络块,其中,i为小于等于I的整数,I为所述超网络模型包括的层级的数量,Ni为正整数;
所述终端根据从所述超网络模型的全部层级中选择的网络块,确定所述目标AI网络模型。
一种实施方式中,所述超网络模型的各个层级对应的Ni可以相同或者不同。
一种实施方式中,所述目标AI网络模型的每一层级对应的Ni等于1。例如:可以从超网络模型的每一个层级中选择一个网络块,即目标AI网络模型的每个层级只有一个网络块。
一种实施方式中,所述Ni个网络块包括第i层级内的全部网络块中权重系数大于预设值的网络块。例如:可以从超网络模型的每一个层级中选择权重系数最大的一个网络块或者权重系数靠前的至少两个网络块。
一种实施方式中,终端可以从超网络模型的每个层级选择至少一个网络块,并根据全部层级中选择的网络块来生成目标AI网络模型。
可选地,终端可以在从超网络模型的每一个层级选择至少一个网络块,并将选择的网络块进行按照层级进行串联后,再根据自身的测量得到的样本数据进行微调,得到目标AI网络模型。
作为一种可选的实施方式,在所述目标AI网络模型中的至少一个层级包括至少两个网络块的情况下,所述方法还包括:
所述终端向所述网络侧设备发送第三信息,其中,所述第三信息包括基于所述超网络模型确定所述目标AI网络模型的方案。
一种实施方式中,基于所述超网络模型确定所述目标AI网络模型,可以是从超网络模型的每一个层级中选择至少一个网络块,并将从超网络模型的全部层级中选择的网络块进行串联,得到目标AI网络模型。此时,基于所述超网络模型确定所述目标AI网络模型的方案可以是如何从超网络模型的每个层级中选择的网络块的方案,或者包括从超网络模型的每个层级中选择哪个或哪些网络块。例如:从所述超网络模型的每个层级中选择的网络块的数量,和/或,选择的网络块是同一层级中权重系数大的和/或是终端能力支持的和/ 或是业务相关的等。
一种实施方式中,基于所述超网络模型确定所述目标AI网络模型的方案可以基于网络块的标识进行描述。例如:第一层级选择的网络块包括{神经网络块3},第二层级选择的网络块包括{神经网络块1,神经网络块3,神经网络块8},第三层级选择的网络块包括{神经网络块4}等等。
一种实施方式中,所述第三信息包括所述目标AI网络模型中的每一个层级的网络块的标识。这样,通过第三信息,网络侧设备可以明确的知道终端实际使用的目标AI网络模型。
本实施方式中,终端通过向网络侧设备发送第三信息,可以使网络侧设备根据第三信息获知终端基于所述超网络模型确定所述目标AI网络模型的方案,进而可以得知终端的目标AI网络模型。
值得提出的是,终端可以在不需要进行模型隐藏的情况下,向网络侧设备发送第三信息,或者,终端默认从超网络模型的每个层级中选择权重系数最大的一个网络块。
当终端需要进行模型隐藏的情况下,终端可以从超网络模型的至少一个层级中选择至少两个网络块,且不向网络侧设备发送第三信息。此时,终端最终确定的目标AI网络模型对于网络侧设备来说是未知的,从而实现了模型隐藏。
例如:用户设备(User Equipment,UE)侧基于超网络模型生成最终要使用的神经网络。一般原则是在每一层级选取权重最大的一个神经网络块,将选出来的所有层级的神经网络快串联起来获得最终的神经网络。但这样的话就不能实现模型隐藏,即UE使用的模型是完全暴露给网络侧的。对此,UE可以根据权重的分布特征在每一层级选取单个或多个神经网络块组成最终的神经网络。比如第一层级选取权重最大的一个神经网络块,第二层级选取权重前三的三个神经网络块进行并联,第三层级选取权重第二大的一个神经网络块,直至选择完所有层级的神经网络块,这样,如果UE不上报自己的模型生成方案(即从超网络生成最终使用的模型的方案),则网络侧无法确切的掌握UE侧使用的AI网络模型。
作为一种可选的实施方式,所述方法还包括:
所述终端向所述网络侧设备发送第四信息,其中,所述第四信息指示所述终端支持的网络块和/或不支持的网络块,所述超网络模型中的网络块为所述终端支持的网络块。
一种实施方式中,终端可以根据自身的硬件和/或软件配置,确定自身能够支持哪些网络块,和/或,确定自身不能够支持哪些网络块。
本实施方式中,终端通过向网络侧设备发送第四信息,可以使网络侧设备据此在确定超网络模型的过程中,仅选择终端支持的网络块来确定超网络模型,这样,可以提升超过网络模型以及基于该超网络模型确定的目标AI网络模型与终端能力的适配。
作为一种可选的实施方式,所述方法还包括:
所述终端向所述网络侧设备发送第五信息,其中,所述第五信息指示所述终端需要的 目标AI网络模型的模型大小和/或复杂度,所述第一AI网络模型为满足所述终端的需求的AI网络模型。
一种实施方式中,上述模型大小可以采用比特进行描述。
可选地,第五信息可以指示终端需要的目标AI网络模型的最大比特数。
一种实施方式中,上述复杂度可以采用浮点数进行描述,如每秒多少个比特数。
可选地,第五信息可以指示终端需要的目标AI网络模型的最大浮点数。
本实施方式中,终端通过向网络侧设备发送第五信息,可以是网络侧设备据此基于超网络模型确定满足终端需求的第一AI网络模型,并将第一AI网络模型下发给终端,终端则可以将第一AI网络模型作为目标AI网络模型,或者在第一AI网络模型的基础上进行微调得到目标AI网络模型。这样,可以实现终端获取的目标AI网络模型的轻量化。
需要说明的是,本申请实施例中终端与网络侧设备之间的信息交互可以是空口信息交互。
可选地,在信息发送端为终端,且信息接收端为网络侧设备的情况下,该交互过程中的信息(如:第一请求信息、第三信息、第四信息、第五信息)可以承载于以下信令或信息中的至少一项:
物理上行控制信道(Physical Uplink Control Channel,PUCCH)的层(layer)1信令;
物理随机接入信道(Physical Random Access Channel,PRACH)的MSG 1;
PRACH的MSG 3;
PRACH的MSG A;
物理上行共享信道(Physical Uplink Shared Channel,PUSCH)承载的信息。
可选地,在信息发送端为网络侧设备,且信息接收端为终端的情况下,该交互过程中的信息(如:第一信息、第二指示信息)可以承载于以下信令或信息中的至少一项:
媒体接入控制控制元素(Medium Access Control Control Element,MAC CE);
无线资源控制(Radio Resource Control,RRC)消息;
非接入层(Non-Access Stratum,NAS)消息;
管理编排消息;
用户面数据;
下行控制信息(Downlink Control Information,DCI)信息;
***信息块(System Information Block,SIB);
物理下行控制信道(Physical Downlink Control Channel,PDCCH)的层1信令;
物理下行共享信道(Physical Downlink Shared Channel,PDSCH)的信息;
PRACH的MSG 2;
PRACH的MSG 4;
PRACH的MSG B。
在本申请实施例中,终端接收来自网络侧设备的第一信息,其中,所述第一信息包括 超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联;所述终端根据所述第一信息确定目标AI网络模型。终端所获取的目标AI网络模型是在超网络模型的基础上确定的,相较于相关技术中,基于训练样本来训练目标AI网络模型的方式而言,能够降低得出目标AI网络模型的计算量。
请参阅图6,本申请实施例提供的信息处理方法,其执行主体可以是网络侧设备,如图6所示,该信息传输方法可以包括以下步骤:
步骤601、网络侧设备向终端发送第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联。
本申请实施例中的第一信息、超网络模型、超网络模型的相关参数、第一AI网络模型、层级、网络块的含义和作用与如图3所示方法实施例中的第一信息、超网络模型、超网络模型的相关参数、第一AI网络模型、层级、网络块的含义和作用相同,在此不作具体限定。
本实施例中的信息传输方法与如图3所示方法实施例相对应,不同之处在于本实施例中的信息传输方法是网络侧设备,而如图3所示方法实施例的执行主体是终端,本实施例中的信息传输方法的解释说明可以参考如图3所示方法实施例中的相关说明,在此不做赘述,且终端和网络侧设备相互配置,例如:网络侧设备确定超网络模型和/或第一AI网络模型,并向终端发送第一信息,以共同实现使终端获取基于超网络模型确定的目标AI网络模型,进而降低终端确定目标AI网络模型的计算量。
作为一种可选的实施方式,所述超网络模型的相关参数,包括以下至少一项:
所述超网络模型中每一个层级具有的网络块的数目;
所述超网络模型中的每一个层级中的每一个网络块的第一参数;
所述超网络模型中的每一个层级中的每一个网络块的权重系数;
第一指示信息,所述第一指示信息指示所述超网络模型中的相邻层之间的连接点的位置或标识。
作为一种可选的实施方式,所述第一指示信息,包括:
所述超网络模型中的第一目标层级内的每一个网络块的输出信息与对应的权重系数的加权求和值,其中,所述超网络模型中的层级包括所述第一目标层级。
作为一种可选的实施方式,所述网络侧设备与所述终端对第二信息达成一致,其中,所述第二信息包括以下至少一项:
所述超网络模型包含的层级的数量;
所述超网络模型的输入信息的格式;
所述超网络模型的输出信息的格式;
网络块集合,所述网络块集合包括所述超网络模型中的全部层级的网络块;
所述网络块集合中的每一个网络块的标识。
作为一种可选的实施方式,在所述网络侧设备向终端发送第一信息之前,所述方法还包括:
所述网络侧设备接收来自所述终端的第一请求信息,其中,所述第一请求信息用于请求所述第一信息。
作为一种可选的实施方式,所述方法还包括:
所述网络侧设备向所述终端发送第二指示信息,其中,所述第二指示信息用于指示所述终端基于所述超网络模型来确定目标AI网络模型。
作为一种可选的实施方式,所述方法还包括:
所述网络侧设备接收来自所述终端的第三信息,其中,所述第三信息包括基于所述超网络模型确定目标AI网络模型的方案,所述目标AI网络模型为所述终端使用的AI网络模型,所述目标AI网络模型中的至少一个层级包括至少两个网络块。
作为一种可选的实施方式,所述第三信息包括所述目标AI网络模型中的每一个层级的网络块的标识。
作为一种可选的实施方式,所述方法还包括:
所述网络侧设备接收来自所述终端的第四信息,其中,所述第四信息指示所述终端支持的网络块和/或不支持的网络块;
所述网络侧设备根据所述第四信息和网络块集合确定所述超网络模型,其中,所述超网络模型中的网络块为所述网络块集合中所述终端支持的网络块。
作为一种可选的实施方式,所述方法还包括:
所述网络侧设备接收来自所述终端的第五信息,其中,所述第五信息指示所述终端需要的目标AI网络模型的模型大小和/或复杂度;
所述网络侧设备根据所述第五信息和所述超网络模型确定所述第一AI网络模型,其中,所述第一AI网络模型为满足所述终端的需求的AI网络模型。
本申请实施例提供的信息传输方法与如图3所述实施中的AI网络模型的确定方法相对应,且能够共同实现使终端获取基于超网络模型确定的目标AI网络模型,进而降低终端确定目标AI网络模型的计算量。
本申请实施例提供的AI网络模型的确定方法,执行主体可以为AI网络模型的确定装置。本申请实施例中以AI网络模型的确定装置执行AI网络模型的确定方法为例,说明本申请实施例提供的AI网络模型的确定装置。
请参阅图7,本申请实施例提供的一种AI网络模型的确定装置,可以是终端内的装置,如图7所示,该AI网络模型的确定装置700可以包括以下模块:
第一接收模块701,用于接收来自网络侧设备的第一信息,其中,所述第一信息包括 超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联;
第一确定模块702,用于根据所述第一信息确定目标AI网络模型。
可选地,所述超网络模型的相关参数,包括以下至少一项:
所述超网络模型中每一个层级具有的网络块的数目;
所述超网络模型中的每一个层级中的每一个网络块的第一参数;
所述超网络模型中的每一个层级中的每一个网络块的权重系数;
第一指示信息,所述第一指示信息指示所述超网络模型中的相邻层级之间的连接点的位置或标识。
可选地,所述第一指示信息,包括:
所述超网络模型中的第一目标层级内的每一个网络块的输出信息与对应的权重系数的加权求和值,其中,所述超网络模型中的层包括所述第一目标层级。
可选地,所述终端与所述网络侧设备对第二信息达成一致,其中,所述第二信息包括以下至少一项:
所述超网络模型包含的层级的数量;
所述超网络模型的输入信息的格式;
所述超网络模型的输出信息的格式;
网络块集合,所述网络块集合包括所述超网络模型中的全部层级的网络块;
所述网络块集合中的每一个网络块的标识。
可选地,AI网络模型的确定装置700还包括:
第二发送模块,用于向所述网络侧设备发送第一请求信息,其中,所述第一请求信息用于请求所述第一信息。
可选地,AI网络模型的确定装置700还包括:
第二接收模块,用于接收来自所述网络侧设备的第二指示信息,其中,所述第二指示信息用于指示所述终端基于所述超网络模型来确定所述目标AI网络模型。
可选地,第一确定模块702,具体用于:
从所述超网络模型的第i个层级中选择Ni个网络块,其中,i为小于等于I的整数,I为所述超网络模型包括的层级的数量,Ni为正整数;
根据从所述超网络模型的全部层级中选择的网络块,确定所述目标AI网络模型。
可选地,所述目标AI网络模型的每一层级对应的Ni等于1。
可选地,所述Ni个网络块包括第i层级内的全部网络块中权重系数大于预设值的网络块。
可选地,在所述目标AI网络模型中的至少一个层级包括至少两个网络块的情况下,AI网络模型的确定装置700还包括:
第三发送模块,用于向所述网络侧设备发送第三信息,其中,所述第三信息包括基于所述超网络模型确定所述目标AI网络模型的方案。
可选地,所述第三信息包括所述目标AI网络模型中的每一个层级的网络块的标识。
可选地,AI网络模型的确定装置700还包括:
第四发送模块,用于向所述网络侧设备发送第四信息,其中,所述第四信息指示所述终端支持的网络块和/或不支持的网络块,所述超网络模型中的网络块为所述终端支持的网络块。
可选地,AI网络模型的确定装置700还包括:
第五发送模块,用于向所述网络侧设备发送第五信息,其中,所述第五信息指示所述终端需要的目标AI网络模型的模型大小和/或复杂度,所述第一AI网络模型为满足所述终端的需求的AI网络模型。
本申请实施例中的AI网络模型的确定装置可以是电子设备,例如具有操作***的电子设备,也可以是电子设备中的部件,例如集成电路或芯片。该电子设备可以是终端,也可以为除终端之外的其他设备。示例性的,终端可以包括但不限于上述所列举的终端11的类型,其他设备可以为服务器、网络附属存储器(Network Attached Storage,NAS)等,本申请实施例不作具体限定。
本申请实施例提供的AI网络模型的确定装置700,能够实现如图3所示方法实施例中终端实现的各个过程,且能够取得相同的有益效果,为避免重复,在此不再赘述。
本申请实施例提供的信息传输方法,执行主体可以为信息传输装置。本申请实施例中以信息传输装置执行信息传输方法为例,说明本申请实施例提供的信息传输装置。
请参阅图8,本申请实施例提供的一种信息传输装置,可以是网络侧设备内的装置,如图8所示,该信息传输装置800可以包括以下模块:
第一发送模块801,用于向终端发送第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联。
可选地,所述超网络模型的相关参数,包括以下至少一项:
所述超网络模型中每一个层级具有的网络块的数目;
所述超网络模型中的每一个层级中的每一个网络块的第一参数;
所述超网络模型中的每一个层级中的每一个网络块的权重系数;
第一指示信息,所述第一指示信息指示所述超网络模型中的相邻层之间的连接点的位置或标识。
可选地,所述第一指示信息,包括:
所述超网络模型中的第一目标层级内的每一个网络块的输出信息与对应的权重系数的加权求和值,其中,所述超网络模型中的层级包括所述第一目标层级。
可选地,所述网络侧设备与所述终端对第二信息达成一致,其中,所述第二信息包括以下至少一项:
所述超网络模型包含的层级的数量;
所述超网络模型的输入信息的格式;
所述超网络模型的输出信息的格式;
网络块集合,所述网络块集合包括所述超网络模型中的全部层级的网络块;
所述网络块集合中的每一个网络块的标识。
可选地,信息传输装置800还包括:
第三接收模块,用于接收来自所述终端的第一请求信息,其中,所述第一请求信息用于请求所述第一信息。
可选地,信息传输装置800还包括:
第六发送模块,用于向所述终端发送第二指示信息,其中,所述第二指示信息用于指示所述终端基于所述超网络模型来确定目标AI网络模型。
可选地,信息传输装置800还包括:
第四接收模块,用于接收来自所述终端的第三信息,其中,所述第三信息包括基于所述超网络模型确定目标AI网络模型的方案,所述目标AI网络模型为所述终端使用的AI网络模型,所述目标AI网络模型中的至少一个层级包括至少两个网络块。
可选地,所述第三信息包括所述目标AI网络模型中的每一个层级的网络块的标识。
可选地,信息传输装置800还包括:
第五接收模块,用于接收来自所述终端的第四信息,其中,所述第四信息指示所述终端支持的网络块和/或不支持的网络块;
第二确定模块,用于根据所述第四信息和网络块集合确定所述超网络模型,其中,所述超网络模型中的网络块为所述网络块集合中所述终端支持的网络块。
可选地,信息传输装置800还包括:
第六接收模块,用于接收来自所述终端的第五信息,其中,所述第五信息指示所述终端需要的目标AI网络模型的模型大小和/或复杂度;
第三确定模块,用于根据所述第五信息和所述超网络模型确定所述第一AI网络模型,其中,所述第一AI网络模型为满足所述终端的需求的AI网络模型。
本申请实施例提供的信息传输装置800,能够实现如图6所示方法实施例中网络侧设备实现的各个过程,且能够取得相同的有益效果,为避免重复,在此不再赘述。
可选地,如图9所示,本申请实施例还提供一种通信设备900,包括处理器901和存储器902,存储器902上存储有可在所述处理器901上运行的程序或指令,例如,该通信设备900为终端时,该程序或指令被处理器901执行时实现如图3所示方法实施例的各个步骤,且能达到相同的技术效果。该通信设备900为网络侧设备时,该程序或指令被处理器901执行时实现如图6所示方法实施例的各个步骤,且能达到相同的技术效果,为避免 重复,这里不再赘述。
本申请实施例还提供一种终端,包括处理器和通信接口,所述通信接口用于接收来自网络侧设备的第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联;所述处理器用于根据所述第一信息确定目标AI网络模型。
该终端实施例能够实现如图7所示AI网络模型的确定装置700执行的各个过程,且能达到相同的技术效果,在此不再赘述。具体地,图10为实现本申请实施例的一种终端的硬件结构示意图。
该终端1000包括但不限于:射频单元1001、网络模块1002、音频输出单元1003、输入单元1004、传感器1005、显示单元1006、用户输入单元1007、接口单元1008、存储器1009以及处理器1010等中的至少部分部件。
本领域技术人员可以理解,终端1000还可以包括给各个部件供电的电源(比如电池),电源可以通过电源管理***与处理器1010逻辑相连,从而通过电源管理***实现管理充电、放电、以及功耗管理等功能。图10中示出的终端结构并不构成对终端的限定,终端可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件布置,在此不再赘述。
应理解的是,本申请实施例中,输入单元1004可以包括图形处理器(Graphics Processing Unit,GPU)10041和麦克风10042,图形处理器10041对在视频捕获模式或图像捕获模式中由图像捕获装置(如摄像头)获得的静态图片或视频的图像数据进行处理。显示单元1006可包括显示面板10061,可以采用液晶显示器、有机发光二极管等形式来配置显示面板10061。用户输入单元1007包括触控面板10071以及其他输入设备10072中的至少一种。触控面板10071,也称为触摸屏。触控面板10071可包括触摸检测装置和触摸控制器两个部分。其他输入设备10072可以包括但不限于物理键盘、功能键(比如音量控制按键、开关按键等)、轨迹球、鼠标、操作杆,在此不再赘述。
本申请实施例中,射频单元1001接收来自网络侧设备的下行数据后,可以传输给处理器1010进行处理;另外,射频单元1001可以向网络侧设备发送上行数据。通常,射频单元1001包括但不限于天线、放大器、收发信机、耦合器、低噪声放大器、双工器等。
存储器1009可用于存储软件程序或指令以及各种数据。存储器1009可主要包括存储程序或指令的第一存储区和存储数据的第二存储区,其中,第一存储区可存储操作***、至少一个功能所需的应用程序或指令(比如声音播放功能、图像播放功能等)等。此外,存储器1009可以包括易失性存储器或非易失性存储器,或者,存储器1009可以包括易失性和非易失性存储器两者。其中,非易失性存储器可以是只读存储器(Read-Only Memory,ROM)、可编程只读存储器(Programmable ROM,PROM)、可擦除可编程只读存储器(Erasable PROM,EPROM)、电可擦除可编程只读存储器(Electrically EPROM,EEPROM) 或闪存。易失性存储器可以是随机存取存储器(Random Access Memory,RAM),静态随机存取存储器(Static RAM,SRAM)、动态随机存取存储器(Dynamic RAM,DRAM)、同步动态随机存取存储器(Synchronous DRAM,SDRAM)、双倍数据速率同步动态随机存取存储器(Double Data Rate SDRAM,DDRSDRAM)、增强型同步动态随机存取存储器(Enhanced SDRAM,ESDRAM)、同步连接动态随机存取存储器(Synch link DRAM,SLDRAM)和直接内存总线随机存取存储器(Direct Rambus RAM,DRRAM)。本申请实施例中的存储器1009包括但不限于这些和任意其它适合类型的存储器。
处理器1010可包括一个或多个处理单元;可选地,处理器1010集成应用处理器和调制解调处理器,其中,应用处理器主要处理涉及操作***、用户界面和应用程序等的操作,调制解调处理器主要处理无线通信信号,如基带处理器。可以理解的是,上述调制解调处理器也可以不集成到处理器1010中。
其中,射频单元1001,用于接收来自网络侧设备的第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联;
处理器1010,用于根据所述第一信息确定目标AI网络模型。
可选地,所述超网络模型的相关参数,包括以下至少一项:
所述超网络模型中每一个层级具有的网络块的数目;
所述超网络模型中的每一个层级中的每一个网络块的第一参数;
所述超网络模型中的每一个层级中的每一个网络块的权重系数;
第一指示信息,所述第一指示信息指示所述超网络模型中的相邻层级之间的连接点的位置或标识。
可选地,所述第一指示信息,包括:
所述超网络模型中的第一目标层级内的每一个网络块的输出信息与对应的权重系数的加权求和值,其中,所述超网络模型中的层包括所述第一目标层级。
可选地,所述终端与所述网络侧设备对第二信息达成一致,其中,所述第二信息包括以下至少一项:
所述超网络模型包含的层级的数量;
所述超网络模型的输入信息的格式;
所述超网络模型的输出信息的格式;
网络块集合,所述网络块集合包括所述超网络模型中的全部层级的网络块;
所述网络块集合中的每一个网络块的标识。
可选地,射频单元1001在执行所述接收来自网络侧设备的第一信息之前,还用于向所述网络侧设备发送第一请求信息,其中,所述第一请求信息用于请求所述第一信息。
可选地,在处理器1010执行所述根据所述第一信息确定目标AI网络模型之前:
射频单元1001,还用于接收来自所述网络侧设备的第二指示信息,其中,所述第二指示信息用于指示所述终端基于所述超网络模型来确定所述目标AI网络模型。
可选地,处理器1010执行的所述根据所述第一信息确定目标AI网络模型,包括:
从所述超网络模型的第i个层级中选择Ni个网络块,其中,i为小于等于I的整数,I为所述超网络模型包括的层级的数量,Ni为正整数;
根据从所述超网络模型的全部层级中选择的网络块,确定所述目标AI网络模型。
可选地,所述目标AI网络模型的每一层级对应的Ni等于1。
可选地,所述Ni个网络块包括第i层级内的全部网络块中权重系数大于预设值的网络块。
可选地,在所述目标AI网络模型中的至少一个层级包括至少两个网络块的情况下:
射频单元1001,还用于向所述网络侧设备发送第三信息,其中,所述第三信息包括基于所述超网络模型确定所述目标AI网络模型的方案。
可选地,所述第三信息包括所述目标AI网络模型中的每一个层级的网络块的标识。
可选地,射频单元1001,还用于向所述网络侧设备发送第四信息,其中,所述第四信息指示所述终端支持的网络块和/或不支持的网络块,所述超网络模型中的网络块为所述终端支持的网络块。
可选地,射频单元1001,还用于向所述网络侧设备发送第五信息,其中,所述第五信息指示所述终端需要的目标AI网络模型的模型大小和/或复杂度,所述第一AI网络模型为满足所述终端的需求的AI网络模型。
本申请实施例提供的终端1000能够实现如图7所示AI网络模型的确定装置700执行的各个过程,且能够取得相同的有益效果,为避免重复,在此不再赘述。
本申请实施例还提供一种网络侧设备,包括处理器和通信接口,所述通信接口用于向终端发送第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联。
该网络侧设备实施例能够实现如图8所示信息传输装置800执行的各个过程,且能达到相同的技术效果,在此不再赘述。具体地,本申请实施例还提供了一种网络侧设备。如图11所示,该网络侧设备1100包括:天线1101、射频装置1102、基带装置1103、处理器1104和存储器1105。天线1101与射频装置1102连接。在上行方向上,射频装置1102通过天线1101接收信息,将接收的信息发送给基带装置1103进行处理。在下行方向上,基带装置1103对要发送的信息进行处理,并发送给射频装置1102,射频装置1102对收到的信息进行处理后经过天线1101发送出去。
以上实施例中网络侧设备执行的方法可以在基带装置1103中实现,该基带装置1103包括基带处理器。
基带装置1103例如可以包括至少一个基带板,该基带板上设置有多个芯片,如图11 所示,其中一个芯片例如为基带处理器,通过总线接口与存储器1105连接,以调用存储器1105中的程序,执行以上方法实施例中所示的网络设备操作。
该网络侧设备还可以包括网络接口1106,该接口例如为通用公共无线接口(Common Public Radio Interface,CPRI)。
具体地,本申请实施例的网络侧设备1100还包括:存储在存储器1105上并可在处理器1104上运行的指令或程序,处理器1104调用存储器1105中的指令或程序执行图8所示各模块执行的方法,并达到相同的技术效果,为避免重复,故不在此赘述。
本申请实施例还提供一种可读存储介质,所述可读存储介质上存储有程序或指令,该程序或指令被处理器执行时实现如图3或图6所示方法实施例的各个过程,且能达到相同的技术效果,为避免重复,这里不再赘述。
其中,所述处理器为上述实施例中所述的终端中的处理器。所述可读存储介质,包括计算机可读存储介质,如计算机只读存储器ROM、随机存取存储器RAM、磁碟或者光盘等。
本申请实施例另提供了一种芯片,所述芯片包括处理器和通信接口,所述通信接口和所述处理器耦合,所述处理器用于运行程序或指令,实现如图3或图6所示方法实施例的各个过程,且能达到相同的技术效果,为避免重复,这里不再赘述。
应理解,本申请实施例提到的芯片还可以称为***级芯片,***芯片,芯片***或片上***芯片等。
本申请实施例另提供了一种计算机程序/程序产品,所述计算机程序/程序产品被存储在存储介质中,所述计算机程序/程序产品被至少一个处理器执行以实现如图3或图6所示方法实施例的各个过程,且能达到相同的技术效果,为避免重复,这里不再赘述。
本申请实施例还提供了一种通信***,包括:终端和网络侧设备,所述终端可用于执行如图3所示的AI网络模型的确定方法的步骤,所述网络侧设备可用于执行如图6所示的信息传输方法的步骤。
需要说明的是,在本文中,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、物品或者装置不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、物品或者装置所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、方法、物品或者装置中还存在另外的相同要素。此外,需要指出的是,本申请实施方式中的方法和装置的范围不限按示出或讨论的顺序来执行功能,还可包括根据所涉及的功能按基本同时的方式或按相反的顺序来执行功能,例如,可以按不同于所描述的次序来执行所描述的方法,并且还可以添加、省去、或组合各种步骤。另外,参照某些示例所描述的特征可在其他示例中被组合。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件,但很多情况下前者 是更佳的实施方式。基于这样的理解,本申请的技术方案本质上或者说对相关技术做出贡献的部分可以以计算机软件产品的形式体现出来,该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中,包括若干指令用以使得一台终端(可以是手机,计算机,服务器,空调器,或者网络设备等)执行本申请各个实施例所述的方法。
上面结合附图对本申请的实施例进行了描述,但是本申请并不局限于上述的具体实施方式,上述的具体实施方式仅仅是示意性的,而不是限制性的,本领域的普通技术人员在本申请的启示下,在不脱离本申请宗旨和权利要求所保护的范围情况下,还可做出很多形式,均属于本申请的保护之内。

Claims (27)

  1. 一种AI网络模型的确定方法,包括:
    终端接收来自网络侧设备的第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联;
    所述终端根据所述第一信息确定目标AI网络模型。
  2. 根据权利要求1所述的方法,其中,所述超网络模型的相关参数,包括以下至少一项:
    所述超网络模型中每一个层级具有的网络块的数目;
    所述超网络模型中的每一个层级中的每一个网络块的第一参数;
    所述超网络模型中的每一个层级中的每一个网络块的权重系数;
    第一指示信息,所述第一指示信息指示所述超网络模型中的相邻层级之间的连接点的位置或标识。
  3. 根据权利要求2所述的方法,其中,所述第一指示信息,包括:
    所述超网络模型中的第一目标层级内的每一个网络块的输出信息与对应的权重系数的加权求和值,其中,所述超网络模型中的层包括所述第一目标层级。
  4. 根据权利要求1所述的方法,其中,所述终端与所述网络侧设备对第二信息达成一致,其中,所述第二信息包括以下至少一项:
    所述超网络模型包含的层级的数量;
    所述超网络模型的输入信息的格式;
    所述超网络模型的输出信息的格式;
    网络块集合,所述网络块集合包括所述超网络模型中的全部层级的网络块;
    所述网络块集合中的每一个网络块的标识。
  5. 根据权利要求1至4中任一项所述的方法,其中,在所述终端接收来自网络侧设备的第一信息之前,所述方法还包括:
    所述终端向所述网络侧设备发送第一请求信息,其中,所述第一请求信息用于请求所述第一信息。
  6. 根据权利要求1至4中任一项所述的方法,其中,在所述终端根据所述第一信息确定目标AI网络模型之前,所述方法还包括:
    所述终端接收来自所述网络侧设备的第二指示信息,其中,所述第二指示信息用于指示所述终端基于所述超网络模型来确定所述目标AI网络模型。
  7. 根据权利要求1至4中任一项所述的方法,其中,所述终端根据所述第一信息确定目标AI网络模型,包括:
    所述终端从所述超网络模型的第i个层级中选择Ni个网络块,其中,i为小于等于I的整数,I为所述超网络模型包括的层级的数量,Ni为正整数;
    所述终端根据从所述超网络模型的全部层级中选择的网络块,确定所述目标AI网络模型。
  8. 根据权利要求7所述的方法,其中,所述目标AI网络模型的每一层级对应的Ni等于1。
  9. 根据权利要求7所述的方法,其中,所述Ni个网络块包括第i层级内的全部网络块中权重系数大于预设值的网络块。
  10. 根据权利要求7所述的方法,其中,在所述目标AI网络模型中的至少一个层级包括至少两个网络块的情况下,所述方法还包括:
    所述终端向所述网络侧设备发送第三信息,其中,所述第三信息包括基于所述超网络模型确定所述目标AI网络模型的方案。
  11. 根据权利要求10所述的方法,其中,所述第三信息包括所述目标AI网络模型中的每一个层级的网络块的标识。
  12. 根据权利要求1至4中任一项所述的方法,其中,所述方法还包括:
    所述终端向所述网络侧设备发送第四信息,其中,所述第四信息指示所述终端支持的网络块和/或不支持的网络块,所述超网络模型中的网络块为所述终端支持的网络块。
  13. 根据权利要求1至4中任一项所述的方法,其中,所述方法还包括:
    所述终端向所述网络侧设备发送第五信息,其中,所述第五信息指示所述终端需要的目标AI网络模型的模型大小和/或复杂度,所述第一AI网络模型为满足所述终端的需求的AI网络模型。
  14. 一种信息传输方法,包括:
    网络侧设备向终端发送第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联。
  15. 根据权利要求14所述的方法,其中,所述超网络模型的相关参数,包括以下至少一项:
    所述超网络模型中每一个层级具有的网络块的数目;
    所述超网络模型中的每一个层级中的每一个网络块的第一参数;
    所述超网络模型中的每一个层级中的每一个网络块的权重系数;
    第一指示信息,所述第一指示信息指示所述超网络模型中的相邻层之间的连接点的位置或标识。
  16. 根据权利要求15所述的方法,其中,所述第一指示信息,包括:
    所述超网络模型中的第一目标层级内的每一个网络块的输出信息与对应的权重系数 的加权求和值,其中,所述超网络模型中的层级包括所述第一目标层级。
  17. 根据权利要求14所述的方法,其中,所述网络侧设备与所述终端对第二信息达成一致,其中,所述第二信息包括以下至少一项:
    所述超网络模型包含的层级的数量;
    所述超网络模型的输入信息的格式;
    所述超网络模型的输出信息的格式;
    网络块集合,所述网络块集合包括所述超网络模型中的全部层级的网络块;
    所述网络块集合中的每一个网络块的标识。
  18. 根据权利要求14至17中任一项所述的方法,其中,在所述网络侧设备向终端发送第一信息之前,所述方法还包括:
    所述网络侧设备接收来自所述终端的第一请求信息,其中,所述第一请求信息用于请求所述第一信息。
  19. 根据权利要求14至17中任一项所述的方法,其中,所述方法还包括:
    所述网络侧设备向所述终端发送第二指示信息,其中,所述第二指示信息用于指示所述终端基于所述超网络模型来确定目标AI网络模型。
  20. 根据权利要求14至17中任一项所述的方法,其中,所述方法还包括:
    所述网络侧设备接收来自所述终端的第三信息,其中,所述第三信息包括基于所述超网络模型确定目标AI网络模型的方案,所述目标AI网络模型为所述终端使用的AI网络模型,所述目标AI网络模型中的至少一个层级包括至少两个网络块。
  21. 根据权利要求20所述的方法,其中,所述第三信息包括所述目标AI网络模型中的每一个层级的网络块的标识。
  22. 根据权利要求14至17中任一项所述的方法,其中,所述方法还包括:
    所述网络侧设备接收来自所述终端的第四信息,其中,所述第四信息指示所述终端支持的网络块和/或不支持的网络块;
    所述网络侧设备根据所述第四信息和网络块集合确定所述超网络模型,其中,所述超网络模型中的网络块为所述网络块集合中所述终端支持的网络块。
  23. 根据权利要求14至17中任一项所述的方法,其中,所述方法还包括:
    所述网络侧设备接收来自所述终端的第五信息,其中,所述第五信息指示所述终端需要的目标AI网络模型的模型大小和/或复杂度;
    所述网络侧设备根据所述第五信息和所述超网络模型确定所述第一AI网络模型,其中,所述第一AI网络模型为满足所述终端的需求的AI网络模型。
  24. 一种AI网络模型的确定装置,应用于终端,所述装置包括:
    第一接收模块,用于接收来自网络侧设备的第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述 多个网络块的至少两个网络块并联;
    第一确定模块,用于根据所述第一信息确定目标AI网络模型。
  25. 一种信息传输装置,应用于网络侧设备,所述装置包括:
    第一发送模块,用于向终端发送第一信息,其中,所述第一信息包括超网络模型的相关参数或基于超网络模型生成的第一AI网络模型的信息,所述超网络模型包括至少一个层级,每一个层级具有至少一个网络块,在同一层级中有多个网络块时,所述多个网络块的至少两个网络块并联。
  26. 一种通信设备,包括处理器和存储器,所述存储器存储可在所述处理器上运行的程序或指令,所述程序或指令被所述处理器执行时实现如权利要求1至13中任一项所述的AI网络模型的确定方法的步骤,或者实现如权利要求14至23中任一项所述的信息传输方法的步骤。
  27. 一种可读存储介质,所述可读存储介质上存储程序或指令,所述程序或指令被处理器执行时实现如权利要求1至13中任一项所述的AI网络模型的确定方法的步骤,或者实现如权利要求14至23中任一项所述的信息传输方法的步骤。
PCT/CN2023/136611 2022-12-07 2023-12-06 Ai网络模型的确定方法、信息传输方法、装置和通信设备 WO2024120409A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211567286.2A CN118158118A (zh) 2022-12-07 2022-12-07 Ai网络模型的确定方法、信息传输方法、装置和通信设备
CN202211567286.2 2022-12-07

Publications (1)

Publication Number Publication Date
WO2024120409A1 true WO2024120409A1 (zh) 2024-06-13

Family

ID=91295521

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/136611 WO2024120409A1 (zh) 2022-12-07 2023-12-06 Ai网络模型的确定方法、信息传输方法、装置和通信设备

Country Status (2)

Country Link
CN (1) CN118158118A (zh)
WO (1) WO2024120409A1 (zh)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113435585A (zh) * 2021-07-15 2021-09-24 支付宝(杭州)信息技术有限公司 一种业务处理方法、装置及设备
CN113570027A (zh) * 2021-06-24 2021-10-29 华为技术有限公司 用于生成神经网络模型的方法、装置、***、设备、介质和程序产品
US20210383240A1 (en) * 2020-06-09 2021-12-09 Toyota Research Institute, Inc. Neural architecture search system for generating a neural network architecture
CN114610272A (zh) * 2020-11-25 2022-06-10 共达地创新技术(深圳)有限公司 Ai模型生成方法、电子设备及存储介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20210383240A1 (en) * 2020-06-09 2021-12-09 Toyota Research Institute, Inc. Neural architecture search system for generating a neural network architecture
CN114610272A (zh) * 2020-11-25 2022-06-10 共达地创新技术(深圳)有限公司 Ai模型生成方法、电子设备及存储介质
CN113570027A (zh) * 2021-06-24 2021-10-29 华为技术有限公司 用于生成神经网络模型的方法、装置、***、设备、介质和程序产品
CN113435585A (zh) * 2021-07-15 2021-09-24 支付宝(杭州)信息技术有限公司 一种业务处理方法、装置及设备

Also Published As

Publication number Publication date
CN118158118A (zh) 2024-06-07

Similar Documents

Publication Publication Date Title
WO2024120409A1 (zh) Ai网络模型的确定方法、信息传输方法、装置和通信设备
WO2023066288A1 (zh) 模型请求方法、模型请求处理方法及相关设备
WO2024120358A1 (zh) 信息传输方法、信息传输装置和通信设备
WO2024067281A1 (zh) Ai模型的处理方法、装置及通信设备
WO2024032694A1 (zh) Csi预测处理方法、装置、通信设备及可读存储介质
WO2024067280A1 (zh) 更新ai模型参数的方法、装置及通信设备
CN115022172A (zh) 信息处理方法、装置、通信设备及可读存储介质
WO2024032695A1 (zh) Csi预测处理方法、装置、通信设备及可读存储介质
Wang et al. Deep transfer reinforcement learning for resource allocation in hybrid multiple access systems
WO2024041421A1 (zh) 测量反馈处理方法、装置、终端及网络侧设备
WO2024067665A1 (zh) Csi预测处理方法、装置、通信设备及可读存储介质
WO2024140512A1 (zh) 数据集确定方法、信息传输方法、装置和通信设备
WO2024131622A1 (zh) Ai网络模型确定方法、信息传输方法、装置和通信设备
WO2024067878A1 (zh) 信道预测方法、装置及通信设备
WO2024093713A1 (zh) 资源配置方法、装置、通信设备及可读存储介质
WO2023179540A1 (zh) 信道预测方法、装置及无线通信设备
WO2024078405A1 (zh) 传输方法、装置、通信设备及可读存储介质
WO2023134628A1 (zh) 传输方法、装置和设备
WO2023125951A1 (zh) 通信模型配置方法、装置和通信设备
WO2024027683A1 (zh) 模型匹配方法、装置、通信设备及可读存储介质
WO2024083004A1 (zh) Ai模型配置方法、终端及网络侧设备
WO2024120445A1 (zh) 模型输入信息的确定方法、装置、设备、***及存储介质
WO2023134761A1 (zh) 波束信息交互方法、装置、设备及存储介质
WO2023098585A1 (zh) 信息传输方法、装置及通信设备
WO2023213270A1 (zh) 模型训练处理方法、装置、终端及网络侧设备