WO2024067280A1 - 更新ai模型参数的方法、装置及通信设备 - Google Patents

更新ai模型参数的方法、装置及通信设备 Download PDF

Info

Publication number
WO2024067280A1
WO2024067280A1 PCT/CN2023/119938 CN2023119938W WO2024067280A1 WO 2024067280 A1 WO2024067280 A1 WO 2024067280A1 CN 2023119938 W CN2023119938 W CN 2023119938W WO 2024067280 A1 WO2024067280 A1 WO 2024067280A1
Authority
WO
WIPO (PCT)
Prior art keywords
parameters
neurons
model
row
order
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2023/119938
Other languages
English (en)
French (fr)
Inventor
杨昂
孙鹏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Vivo Mobile Communication Co Ltd
Original Assignee
Vivo Mobile Communication Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Vivo Mobile Communication Co Ltd filed Critical Vivo Mobile Communication Co Ltd
Publication of WO2024067280A1 publication Critical patent/WO2024067280A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00Machine learning
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/0499Feedforward networks
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • G06N3/084Backpropagation, e.g. using gradient descent
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • G06N3/098Distributed learning, e.g. federated learning
    • 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
    • H04L41/08Configuration management of networks or network elements
    • H04L41/0803Configuration setting
    • H04L41/0813Configuration setting characterised by the conditions triggering a change of settings
    • H04L41/082Configuration setting characterised by the conditions triggering a change of settings the condition being updates or upgrades of network functionality

Definitions

  • the present application belongs to the field of communication technology, and specifically relates to a method, device and communication equipment for updating AI model parameters.
  • AI Artificial Intelligence
  • neural networks such as neural networks, decision trees, support vector machines, Bayesian classifiers, etc.
  • the model is updated by transmitting the AI model.
  • this update method is inefficient and has a large network signaling overhead.
  • the embodiments of the present application provide a method, device and communication equipment for updating AI model parameters, which solve the problems of low efficiency of AI model update method and high network signaling overhead in related technologies.
  • a method for updating AI model parameters comprising:
  • the first node sends first information related to the AI model parameters to the second node;
  • the first information includes at least one of the following: an update mode of the AI model parameters, and an indication method of the update parameters of the AI model parameters.
  • a method for updating AI model parameters comprising:
  • the second node receives first information related to AI model parameters sent by the first node
  • the second node updates the AI model parameters according to the first information
  • the first information includes at least one of the following: an update mode of the AI model parameters, and an indication method of the update parameters of the AI model parameters.
  • a device for updating AI model parameters comprising:
  • a first sending module configured to send first information related to the AI model parameters to the second node
  • the first information includes at least one of the following: an update mode of the AI model parameters, and an indication method of the update parameters of the AI model parameters.
  • a device for updating AI model parameters comprising:
  • a first receiving module configured to receive first information related to AI model parameters sent by the first node
  • An updating module configured to update the AI model parameters according to the first information
  • the first information includes at least one of the following: an update mode of the AI model parameters, and an indication method of the update parameters of the AI model parameters.
  • a communication device comprising: a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the method described in the first aspect or the second 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 or the second 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 steps of the method described in the first aspect or the second aspect.
  • a computer program/program product is provided, wherein the computer program/program product is stored in a non-volatile storage medium, and the program/program product is executed by at least one processor to implement the steps of the method described in the first aspect or the second aspect.
  • a communication system comprising a terminal and a network side device, the terminal being used to execute the steps of the method described in the first aspect, and the network side device being used to execute the steps of the method described in the second aspect.
  • the first node may send the update mode of the AI model parameters and/or the indication method of the update parameters of the AI model parameters to the second node.
  • the second node does not need to compile or recompile the AI model.
  • the second node updates the existing AI model parameters according to the received update mode of the AI model parameters and/or the indication method of the update parameters of the AI model parameters, which can effectively improve the efficiency of transmitting the AI model in the wireless communication system and reduce network signaling overhead.
  • Figure 1 is a schematic diagram of a neural network
  • Figure 2 is a schematic diagram of a neuron
  • FIG3 is a schematic diagram of the architecture of a wireless communication system according to an embodiment of the present application.
  • FIG4 is a flowchart of a method for updating AI model parameters according to an embodiment of the present application.
  • FIG5 is a second flowchart of a method for updating AI model parameters according to an embodiment of the present application.
  • FIG6 is a schematic diagram of one of the indication methods of AI model parameters according to an embodiment of the present application.
  • FIG. 7 is a second schematic diagram of the indication method of AI model parameters according to an embodiment of the present application.
  • FIG8 is a third schematic diagram of the indication method of AI model parameters according to an embodiment of the present application.
  • FIG9 is a fourth schematic diagram of the indication method of AI model parameters according to an embodiment of the present application.
  • FIG10 is a structural diagram of a device for updating AI model parameters according to an embodiment of the present application.
  • FIG11 is a second structural diagram of the device for updating AI model parameters according to an embodiment of the present application.
  • FIG12 is a schematic diagram of a terminal according to an embodiment of the present application.
  • FIG13 is a schematic diagram of a network side device according to an embodiment of the present application.
  • FIG14 is a schematic diagram of a communication device according to 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
  • 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
  • This application uses a neural network as an example for illustration, but does not limit the specific type of AI module.
  • the structure of the neural network is shown in FIG1 .
  • the neural network is composed of neurons, and the schematic diagram of neurons is shown in Figure 2.
  • a 1 , a 2 , ... a K are inputs
  • w is the weight (multiplicative coefficient)
  • b is the bias (additive coefficient)
  • ⁇ (.) is the activation function
  • z a 1 *w 1 + ... + a k *w k + ... + a K *w K + b.
  • Common activation functions include Sigmoid function, tanh function, Rectified Linear Unit (ReLU), etc.
  • the parameters of a neural network can be optimized using an optimization algorithm.
  • An optimization algorithm is a type of algorithm that can minimize or maximize an objective function (sometimes called a 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, a neural network model f(.) is constructed. Once the model is in place, Based on the input x, we can get the predicted output f(x), and we can calculate the difference between the predicted value and the true value (f(x)-Y), which is the loss function. If we find a suitable W,b to minimize the value of the above loss function, the smaller the loss value, the closer the model is to the actual situation.
  • the common optimization algorithms are basically based on the error back propagation (BP) algorithm.
  • BP error 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 error back propagation stage.
  • Error back propagation 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 in 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 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 pre-set number of learning times is reached.
  • the selected AI algorithms and models vary depending on the type of solution.
  • the main way to improve the network performance of the fifth generation mobile communication technology (5th Generation, 5G) with the help of AI is to enhance or replace existing algorithms or processing modules through algorithms and models based on neural networks.
  • algorithms and models based on neural networks can achieve better performance than those based on deterministic algorithms.
  • the more commonly used neural networks include deep neural networks, convolutional neural networks, and recurrent neural networks.
  • FIG3 shows a block diagram of a wireless communication system that can be applied to an embodiment of the present application.
  • the wireless communication system includes a terminal 31 and a network side device 32.
  • the terminal 31 may be a mobile phone, a tablet computer (Tablet Personal Computer), a laptop computer (Laptop Computer) or a notebook computer, a personal digital assistant (PDA), a handheld computer, a netbook, an ultra-mobile personal computer (UMPC), a mobile Internet device (Mobile Internet Device, MID), an augmented reality (AR)/virtual reality (VR) device, a robot, a wearable device (Wearable Device), a vehicle user equipment (VUE), a pedestrian terminal (Pedestrian User Equipment, PUE), a smart home (a home appliance with wireless communication function, such as a refrigerator, a television, a washing machine or furniture, etc.), a game console, a personal computer (PC), a teller machine or a self-service machine and other terminal side devices, and the wearable devices include: a smart watch, a smart bracelet, a smart Headphones, smart glasses, smart jewelry (smart bracelets, smart bracelets, smart rings, smart necklaces, smart anklets, smart anklets, etc
  • the terminal involved in this application can also be a chip in the terminal, such as a modem chip, a system-on-chip (SoC). It should be noted that the specific type of the terminal 31 is not limited in the embodiment of this application.
  • the network side device 32 may include an access network device or a core network device, wherein the access network device may also be referred to as a wireless access network device, a wireless access network (Radio Access Network, RAN), a wireless access network function or a wireless access network unit.
  • the access network device may include a base station, a wireless local area network (Wireless Local Area Network, WLAN) access point or a WiFi node, etc.
  • the base station may be referred to as a node B, an evolved node B (eNB), an access point, a base transceiver station (Base Transceiver Station, BTS), a radio base station, a radio transceiver, a basic service set (Basic Service Set, BSS), an extended service set (Extended Service Set, ESS), a home B node, a home evolved B node, a transmitting and receiving point (Transmitting Receiving Point, TRP) or other appropriate terms in the field, as long as the same technical effect is achieved, the base station is not limited to a specific technical vocabulary, it should be noted that in the embodiment of the present application, only the base station in the NR system is used as an example for introduction, and the specific type of the base station is not limited.
  • the core network equipment may include but is not limited to at least one of the following: core network nodes, core network functions, mobility management entity (Mobility Management Entity, MME), access and mobility management function (Access and Mobility Management Function, AMF), session management function (Session Management Function, SMF), user plane function (User Plane Function, UPF), policy control function (Policy Control Function, PCF), policy and charging rules function unit (Policy and Charging Rules Function, PCRF), edge application service discovery function (Edge Application Server Discovery Function, EASDF), unified data management (Unifi).
  • MME mobility management entity
  • AMF Access and Mobility Management Function
  • SMF Session Management Function
  • SMF session management function
  • User Plane Function User Plane Function
  • Policy Control Function Policy Control Function
  • PCRF Policy and Charging Rules Function
  • EASDF edge application service discovery function
  • EASDF unified data management
  • an embodiment of the present application provides a method for updating AI model parameters, which is applied to a first node, which may be a network side device or a terminal.
  • the specific steps include: Step 401.
  • Step 401 The first node sends first information related to AI model parameters to the second node;
  • the first information includes at least one of the following: an update mode of the AI model parameters, and an indication method of the AI model parameters.
  • the first information is used by the second node to update the AI model parameters.
  • the update mode of the AI model parameters in the first information is used to indicate the method used to update the AI model parameters.
  • the indication method is used to indicate the location of the update parameter or to indicate the sending order of the AI model parameters.
  • the first node can indicate the sending order of all AI model parameters to the second node, and the second node updates the AI model according to some of all the received AI model parameters according to the situation; or the first node can indicate the sending order of some AI model parameters to the second node, and the second node updates the AI model according to the received some AI model parameters, where some AI model parameters can also be called AI model update parameters.
  • the first node can send the update mode of the AI model parameters and/or the indication method of the AI model parameters to the second node.
  • the second node does not need to compile or recompile the AI model.
  • the second node updates the existing AI model parameters according to the received update mode of the AI model parameters and/or the indication method of the AI model parameters, which can effectively improve the efficiency of transmitting the AI model in the wireless communication system and reduce network signaling overhead.
  • the first node is a first network side device or a first terminal
  • the second node is a second network side device or a second terminal.
  • the first node is a first network side device and the second node is a second terminal
  • the first node is a first terminal and the second node is a second network side device
  • the first node is a first network side device and the second node is a second network side device
  • the first node is a first terminal and the second node is a second terminal
  • the first node is a first terminal and the second node is a second terminal.
  • the update mode includes one of the following:
  • the first mode includes: updating all parameters in a first parameter subset, where the first parameter subset is a subset of the AI model parameters, and the first parameter subset is pre-configured or agreed upon by a protocol or indicated by a network side;
  • the second mode includes: updating all parameters in a second parameter subset, where the second parameter subset is a subset of the first parameter subset;
  • the third mode includes: updating all parameters of the AI model;
  • the fourth mode includes: updating a third parameter subset, wherein the third parameter subset is a subset of the AI model parameters.
  • the indication method includes one of the following:
  • first information where the first information is used to indicate the location of the update parameter in the AI model
  • Second information where the second information is used to indicate the order in which updated parameters in the AI model are sent.
  • FIG1 the structure of a neural network applicable to an embodiment of the present application is shown in FIG1 , where one layer corresponds to one column, i.e., the input layer is the first column, the output layer is the last column, and the hidden layer is the middle column.
  • the embodiment of the present application may also be applicable to neural networks of other structures.
  • the structure of the neural network may be another structure of the neural network obtained by rotating according to FIG. 1 , so that another structure based on the neural network “indicates the position of the update parameter in the AI model”, and “Indicating the order of sending updated parameters in the AI model”, the indication method is applied to the structure of the neural network shown in FIG1 , and is similar to other structures applied to the neural network, and will not be repeated here.
  • another structure of the neural network can be rotated 90 degrees to the right relative to the schematic diagram of Figure 1, that is, one layer corresponds to one row, the input layer is the first row, the output layer is the last row, and the hidden layer is the middle row; for another example, another structure of the neural network can be rotated 90 degrees to the left relative to the neural network structure of Figure 1, that is, one layer corresponds to one row, the input layer is the last row, the output layer is the first row, and the hidden layer is the middle row. It can be understood that the structure of the neural network is not limited to the above three forms.
  • the AI model is a neural network, which includes an input layer, one or more hidden layers and an output layer.
  • the input layer, one or more hidden layers, the output layer, and the neurons in each layer form a structure with multiple rows and columns.
  • the order in which the parameters of the neurons in the neural network are sent is determined according to the position of the neurons in the neural network.
  • the order of sending the parameters of the neurons in the neural network includes one of the following:
  • the first column corresponds to the input layer, and the last column corresponds to the output layer; or, the first column corresponds to the output layer, and the last column corresponds to the input layer.
  • the first row corresponds to the input layer, and the last row corresponds to the output layer; or, the first row corresponds to the output layer, and the last row corresponds to the input layer;
  • the first row is the top row in the neural network, and the last row is the bottom row in the neural network; or, the first row is the bottom row in the neural network, and the last row is the top row in the neural network.
  • the order of sending the parameters of neurons in the neural network includes: the parameters of all neurons in the first column, the parameters of all neurons in the second column, and so on, until the parameters of all neurons in the last column, the order of sending the parameters of neurons in the neural network also includes one of the following:
  • the order of sending the parameters of neurons in the neural network when the order of sending the parameters of neurons in the neural network includes the parameters of all neurons in the first row, the parameters of all neurons in the second row, and so on, until the parameters of all neurons in the last row, the order of sending the parameters of neurons in the neural network also includes one of the following:
  • the parameters of the neurons are sent through at least one of the following:
  • Each row sends the parameters of the neurons in the row with the largest number of neurons, and the layers with missing neurons are padded with preset values (e.g., 0).
  • the second and fourth layers of the 50th row will be filled with 0 when sending.
  • Each row sends the parameters of the neuron according to the actual number of neurons.
  • the first layer there are 50 neurons in the first layer, 40 neurons in the second layer, 60 neurons in the third layer, and 30 neurons in the fourth layer. If there are no neurons in the second and fourth layers in line 50, the second and fourth layers are skipped when sending, and the parameters of the neurons in the first and third layers are sent in line 50.
  • the sending order of the AI model parameters is preconfigured or agreed upon by the protocol or indicated by the network side.
  • the AI model parameters include multiplicative coefficients and additive coefficients, and the sending priority of the multiplicative coefficients is higher than the sending priority of the additive coefficients; or, the sending priority of the additive coefficients is higher than the sending priority of the multiplicative coefficients.
  • the quantization level and/or compression method of the AI model is pre-configured or agreed upon by the protocol or indicated by the network side.
  • the model identification (Identity, ID) of the updated AI model is the same as the model ID of the AI model before the update, and the AI model before the update is not saved;
  • the model ID of the updated AI model is different from the model ID of the AI model before the update, and the AI model before the update is saved, and the model ID of the AI model before the update remains unchanged.
  • the AI model includes a first functional module, and the first functional module is used for at least one of the following:
  • signal processing including but not limited to at least one of the following: signal detection, filtering, equalization, etc.
  • the signal includes but is not limited to at least one of the following: demodulation reference signal (DMRS), sounding reference signal (SRS), synchronization signal block (Synchronization Signal and PBCH block, SSB), tracking reference signal (TRS), phase tracking reference signals (PTRS), channel state information reference signal (CSI-RS), etc.;
  • DMRS demodulation reference signal
  • SRS sounding reference signal
  • SSB synchronization signal block
  • TRS tracking reference signal
  • PTRS phase tracking reference signals
  • CSI-RS channel state information reference signal
  • Channel signal transmission channel signal reception, channel demodulation or channel signal transmission, wherein the channel includes but is not limited to at least one of the following: Physical Downlink Control Channel (PDCCH), Physical Downlink Shared Channel (PDSCH), Physical Uplink Control Channel (PUCCH), Physical Uplink Shared Channel (PUSCH), Physical Random Access Channel (PRACH), Physical Broadcast Channel (PBCH);
  • PDCCH Physical Downlink Control Channel
  • PDSCH Physical Downlink Shared Channel
  • PUCH Physical Uplink Control Channel
  • PUSCH Physical Uplink Shared Channel
  • PRACH Physical Broadcast Channel
  • PBCH Physical Broadcast Channel
  • channel state information feedback includes but is not limited to at least one of the following: channel-related information, channel matrix-related information, channel characteristic information, channel matrix characteristic information, precoding matrix indicator (PMI), rank indicator (RI), CSI-RS resource indicator (CSI-RS Resource Indicator, CRI), channel quality indicator (CQI), layer indicator (LI), etc.
  • PMI precoding matrix indicator
  • RI rank indicator
  • CSI-RS resource indicator CSI-RS Resource Indicator
  • CQI channel quality indicator
  • LI layer indicator
  • Another example is the partial reciprocity of uplink and downlink in Frequency Division Duplex (FDD).
  • FDD Frequency Division Duplex
  • the base station obtains angle and delay information based on the uplink channel, and can notify the terminal of the angle and delay information through CSI-RS precoding or direct indication.
  • the terminal reports according to the indication of the base station or selects and reports within the indication range of the base station, thereby reducing the terminal's calculation workload and the CSI reporting overhead.
  • beam management including but not limited to at least one of the following: beam measurement, beam reporting, beam prediction, beam failure detection, beam failure recovery, and new beam indication during beam failure recovery;
  • channel prediction including but not limited to at least one of the following: prediction of channel state information and beam prediction;
  • Interference suppression including but not limited to at least one of the following: intra-cell interference, inter-cell interference, out-of-band interference, and intermodulation interference;
  • Positioning such as estimating the specific position (including horizontal position and/or vertical position) or possible future trajectory of the terminal through a reference signal (such as SRS), or estimating information of auxiliary position estimation or trajectory estimation of the terminal;
  • predicting or managing high-level services and/or high-level parameters including but not limited to at least one of the following: throughput, required packet size, service demand, mobile speed, noise information, etc.;
  • control signaling including but not limited to at least one of the following: power control related signaling, beam management related signaling.
  • the efficiency of transmitting AI models in wireless communication systems can be effectively improved, and network signaling overhead can be reduced.
  • an embodiment of the present application provides a method for updating AI model parameters, which is applied to a second node, where the second node is a network side device or terminal.
  • the method includes: step 501 and step 502.
  • Step 501 The second node receives first information related to AI model parameters sent by the first node;
  • Step 502 The second node updates the AI model parameters according to the first information
  • the first information includes at least one of the following: an update mode of the AI model parameters, and an indication method of the AI model parameters.
  • the first node is a first network side device or a first terminal
  • the second node is a second network side device or a second terminal.
  • the first node is a first network side device
  • the second node is a second terminal. end, or, the first node is the first terminal and the second node is the second network side device, or, the first node is the first network side device and the second node is the second network side device, or the first node is the first terminal and the second node is the second terminal.
  • the update mode includes one of the following:
  • the first mode includes: updating all parameters in a first parameter subset, where the first parameter subset is a subset of the AI model parameters, and the first parameter subset is pre-configured or agreed upon by a protocol or indicated by a network side;
  • the second mode includes: updating all parameters in a second parameter subset, the second parameter subset being a subset of the first parameter subset;
  • the third mode includes: updating all parameters of the AI model;
  • the fourth mode includes: updating a third parameter subset, wherein the third parameter subset is a subset of the AI model parameters.
  • the indication method includes one of the following:
  • first information where the first information is used to indicate the position of the AI model parameter in the AI model
  • Second information where the second information is used to indicate the order in which updated parameters in the AI model are sent.
  • the AI model is a neural network, which includes an input layer, one or more hidden layers and an output layer.
  • the input layer, one or more hidden layers, the output layer, and the neurons in each layer form a multi-row and multi-column structure, and the order in which the parameters of each neuron in the neural network are sent is determined according to the position of the neuron in the neural network.
  • the order in which the parameters of each neuron in the neural network are sent includes:
  • the first column contains the parameters of all neurons
  • the second column contains the parameters of all neurons, and so on, until the last column contains the parameters of all neurons;
  • the first row contains the parameters of all neurons
  • the second row contains the parameters of all neurons
  • so on until the last row contains the parameters of all neurons.
  • the first column corresponds to the input layer, and the last column corresponds to the output layer; or, the first column corresponds to the output layer, and the last column corresponds to the input layer;
  • the first row corresponds to the input layer, and the last row corresponds to the output layer; or, the first row corresponds to the output layer, and the last row corresponds to the input layer.
  • the parameters of the neurons in each column are sent in the order of neurons from top to bottom, or the parameters of the neurons in each column are sent in the order of neurons from bottom to top, or the sending order of the parameters of the neurons in each column is the same, or the sending order of the parameters of the neurons in adjacent columns is opposite.
  • the first row is the top row in the neural network, and the last row is the bottom row in the neural network; or, the first row is the bottom row in the neural network, and the last row is the top row in the neural network.
  • the parameters of the neurons in each row are sent in the order of neurons from left to right, or the parameters of the neurons in each row are sent in the order of neurons from right to left, or the sending order of the parameters of the neurons in each row is the same, or the sending order of the parameters of the neurons in adjacent rows is opposite.
  • the parameters of the neurons are sent in each row according to the row with the largest number of neurons, and the layers lacking neurons are supplemented with preset values, or the parameters of the neurons are sent in each row according to the actual number of neurons.
  • the sending order of the AI model parameters is preconfigured or agreed upon by the protocol or indicated by the network side.
  • the AI model parameters include multiplicative coefficients and additive coefficients, and the sending priority of the multiplicative coefficients in the AI model is higher than the sending priority of the additive coefficients; or, the sending priority of the additive coefficients in the AI model is higher than the sending priority of the multiplicative coefficients.
  • the quantization level and/or compression method of the AI model is pre-configured or agreed upon by the protocol or indicated by the network side.
  • the model ID of the updated AI model is the same as the model ID of the AI model before the update, and the AI model before the update is not saved;
  • the model ID of the updated AI model is different from the model ID of the AI model before the update, and the AI model before the update is saved, and the model ID of the AI model before the update remains unchanged.
  • the AI model includes a first functional module, and the first functional module is used for at least one of the following:
  • signal processing including but not limited to at least one of the following: signal detection, filtering, equalization, etc., wherein the signal includes but is not limited to at least one of the following: DMRS, SRS, SSB, TRS, PTRS, CSI-RS, etc.;
  • channel signal transmission channel signal reception, channel demodulation or channel signal transmission, where the channel includes but is not limited to at least one of the following: PDCCH, PDSCH, PUCCH, PUSCH, PRACH, PBCH;
  • channel state information feedback includes but is not limited to at least one of the following: channel-related information, channel matrix-related information, channel characteristic information, channel matrix characteristic information, PMI, RI, CRI, CQI, LI, etc.
  • Another example is the partial reciprocity of FDD uplink and downlink.
  • the base station obtains angle and delay information based on the uplink channel, and can notify the terminal of the angle and delay information through CSI-RS precoding or direct indication.
  • the terminal reports according to the indication of the base station or selects and reports within the indication range of the base station, thereby reducing the terminal's calculation workload and the CSI reporting overhead.
  • Beam management including but not limited to at least one of the following: beam measurement, beam reporting, beam prediction, Beam failure detection, beam failure recovery, new beam indication in beam failure recovery;
  • channel prediction including but not limited to at least one of the following: prediction of channel state information and beam prediction;
  • Interference suppression including but not limited to at least one of the following: intra-cell interference, inter-cell interference, out-of-band interference, and intermodulation interference;
  • Positioning such as estimating the specific position (including horizontal position and/or vertical position) or possible future trajectory of the terminal through a reference signal (such as SRS), or estimating information of auxiliary position estimation or trajectory estimation of the terminal;
  • predicting or managing high-level services and/or high-level parameters including but not limited to at least one of the following: throughput, required packet size, service demand, mobile speed, noise information, etc.;
  • control signaling including but not limited to at least one of the following: power control related signaling, beam management related signaling.
  • the efficiency of transmitting AI models in wireless communication systems can be effectively improved, and network signaling overhead can be reduced.
  • an embodiment of the present application provides a device for updating AI model parameters, which is applied to a first node.
  • the device 1000 includes:
  • the first information includes at least one of the following: an update mode of the AI model parameters, and an indication method of the AI model parameters.
  • the update mode includes one of the following: a first mode, a second mode, a third mode, and a fourth mode;
  • the first mode includes: updating all parameters in a first parameter subset, where the first parameter subset is a subset of the AI model parameters, and the first parameter subset is pre-configured or agreed upon by a protocol or indicated by a network side;
  • the second mode includes: updating all parameters in a second parameter subset, the second parameter subset being a subset of the first parameter subset;
  • the third mode includes: updating all parameters of the AI model
  • the fourth mode includes: updating a third parameter subset, where the third parameter subset is a subset of the AI model parameters.
  • the indication method includes one of the following:
  • first information where the first information is used to indicate a position of the AI model parameter in the AI model
  • Second information where the second information is used to indicate the order in which the AI model parameters are sent.
  • the AI model is a neural network, which includes an input layer, one or more hidden layers and an output layer.
  • the input layer, one or more hidden layers, the output layer, and the neurons in each layer form a multi-row and multi-column structure, and the order in which the parameters of each neuron in the neural network are sent is determined according to the position of the neuron in the neural network.
  • the order in which the parameters of each neuron in the neural network are sent includes:
  • the first column contains the parameters of all neurons
  • the second column contains the parameters of all neurons, and so on, until the last column contains all neurons.
  • the parameters of the neuron or,
  • the first row contains the parameters of all neurons
  • the second row contains the parameters of all neurons
  • so on until the last row contains the parameters of all neurons.
  • the first column corresponds to the input layer
  • the last column corresponds to the output layer, or the first column corresponds to the output layer, and the last column corresponds to the input layer
  • the first row corresponds to the input layer and the last row corresponds to the output layer, or the first row corresponds to the output layer and the last row corresponds to the input layer.
  • the order of sending the parameters of neurons in the neural network in the case where the order of sending the parameters of neurons in the neural network includes the parameters of all neurons in the first column, the parameters of all neurons in the second column, and so on, until the parameters of all neurons in the last column, the order of sending the parameters of neurons in the neural network also includes one of the following:
  • the parameters of the neurons in each column are sent in the order of the neurons from top to bottom;
  • the parameters of the neurons in each column are sent in the order of the neurons from bottom to top;
  • the parameters of neurons in adjacent columns are sent in the opposite order.
  • the first row is the top row in the neural network, and the last row is the bottom row in the neural network; or, the first row is the bottom row in the neural network, and the last row is the top row in the neural network.
  • the order of sending the parameters of the neurons in the neural network in the case where the order of sending the parameters of the neurons in the neural network includes the parameters of all neurons in the first row, the parameters of all neurons in the second row, and so on, until the parameters of all neurons in the last row, the order of sending the parameters of the neurons in the neural network also includes one of the following:
  • the parameters of the neurons in each row are sent in the order of neurons from left to right;
  • the parameters of the neurons in each row are sent in the order of neurons from right to left;
  • the parameters of neurons in each row are sent in the same order;
  • the parameters of neurons in adjacent rows are sent in the opposite order.
  • the sending order of the AI model parameters is preconfigured or agreed upon by the protocol or indicated by the network side.
  • the AI model parameters include multiplicative coefficients and additive coefficients, and the sending priority of the multiplicative coefficients is higher than the sending priority of the additive coefficients; or, the sending priority of the additive coefficients is higher than the sending priority of the multiplicative coefficients.
  • the quantization level and/or compression method of the AI model is pre-configured or agreed upon by the protocol or indicated by the network side.
  • the model ID of the updated AI model is the same as the model ID of the AI model before the update, and the AI model before the update is not saved;
  • the model ID of the updated AI model is different from the model ID of the AI model before the update, and the AI model before the update is saved, and the model ID of the AI model before the update remains unchanged.
  • the AI model includes a first functional module, and the first functional module is used for one or more of the following:
  • Forecast or manage high-level business and/or high-level parameter forecasts
  • the first node is a first network-side device or a first terminal
  • the second node is a second network-side device or a second terminal.
  • the device provided in the embodiment of the present application can implement each process implemented by the method embodiment of Figure 4 and achieve the same technical effect. To avoid repetition, it will not be repeated here.
  • an embodiment of the present application provides a device for updating AI model parameters, which is applied to a second node.
  • the device 1100 includes:
  • a first receiving module 1101 is used to receive first information related to AI model parameters sent by a first node
  • An updating module 1102 configured to update the AI model parameters according to the first information
  • the first information includes at least one of the following: an update mode of the AI model parameters, and an indication method of the AI model parameters.
  • the update mode includes one of the following: a first mode, a second mode, a third mode, and a fourth mode;
  • the first mode includes: updating all parameters in a first parameter subset, where the first parameter subset is a subset of the AI model parameters, and the first parameter subset is pre-configured or agreed upon by a protocol or indicated by a network side;
  • the second mode includes: updating all parameters in a second parameter subset, the second parameter subset is a subset of the first parameter subset, the first parameter subset is a subset of the AI model parameters, and the first parameter subset is pre-indicated or agreed upon by protocol;
  • the third mode includes: updating all parameters of the AI model
  • the fourth mode includes: updating a third parameter subset, where the third parameter subset is a subset of the AI model parameters.
  • the indication method includes one of the following:
  • first information where the first information is used to indicate a position of the AI model parameter in the AI model
  • Second information where the second information is used to indicate the order in which the AI model parameters are sent.
  • the AI model is a neural network, which includes an input layer, one or more hidden layers and an output layer.
  • the input layer, one or more hidden layers, the output layer, and the neurons in each layer form a multi-row and multi-column structure, and the order in which the parameters of each neuron in the neural network are sent is determined according to the position of the neuron in the neural network.
  • the order in which the parameters of each neuron in the neural network are sent includes:
  • the first column contains the parameters of all neurons
  • the second column contains the parameters of all neurons, and so on, until the last column contains the parameters of all neurons;
  • the first row contains the parameters of all neurons
  • the second row contains the parameters of all neurons
  • so on until the last row contains the parameters of all neurons.
  • the first column corresponds to the input layer
  • the last column corresponds to the output layer, or the first column corresponds to the output layer, and the last column corresponds to the input layer
  • the first row corresponds to the input layer and the last row corresponds to the output layer, or the first row corresponds to the output layer and the last row corresponds to the input layer.
  • the order of sending the parameters of neurons in the neural network in the case where the order of sending the parameters of neurons in the neural network includes the parameters of all neurons in the first column, the parameters of all neurons in the second column, and so on, until the parameters of all neurons in the last column, the order of sending the parameters of neurons in the neural network also includes one of the following:
  • the parameters of the neurons in each column are sent in the order of the neurons from top to bottom;
  • the parameters of the neurons in each column are sent in the order of the neurons from bottom to top;
  • the parameters of neurons in adjacent columns are sent in the opposite order.
  • the first row is the top row in the neural network, and the last row is the bottom row in the neural network; or, the first row is the bottom row in the neural network, and the last row is the top row in the neural network.
  • the order of sending the parameters of the neurons in the neural network in the case where the order of sending the parameters of the neurons in the neural network includes the parameters of all neurons in the first row, the parameters of all neurons in the second row, and so on, until the parameters of all neurons in the last row, the order of sending the parameters of the neurons in the neural network also includes one of the following:
  • the parameters of the neurons in each row are sent in the order of neurons from left to right;
  • the parameters of the neurons in each row are sent in the order of neurons from right to left;
  • the parameters of neurons in each row are sent in the same order;
  • the parameters of neurons in adjacent rows are sent in the opposite order.
  • each row when the number of neurons in each layer is not exactly the same, each row sends the parameters of the neurons according to the row with the largest number of neurons, and the layers lacking neurons are supplemented with preset values, or each row The parameters of the neurons are sent according to the actual number of neurons.
  • the sending order of the AI model parameters is preconfigured or agreed upon by the protocol or indicated by the network side.
  • the AI model parameters include multiplicative coefficients and additive coefficients, and the sending priority of the multiplicative coefficients is higher than the sending priority of the additive coefficients; or, the sending priority of the additive coefficients is higher than the sending priority of the multiplicative coefficients.
  • the quantization level and/or compression method of the AI model is pre-configured or agreed upon by the protocol or indicated by the network side.
  • the model ID of the updated AI model is the same as the model ID of the AI model before the update, and the AI model before the update is not saved;
  • the model ID of the updated AI model is different from the model ID of the AI model before the update, and the AI model before the update is saved, and the model ID of the AI model before the update remains unchanged.
  • the AI model includes a first functional module, and the first functional module is used for one or more of the following:
  • Forecast or manage high-level business and/or high-level parameter forecasts
  • the second node is a second network-side device or a second terminal
  • the first node is a first network-side device or a first terminal
  • the device provided in the embodiment of the present application can implement each process implemented by the method embodiment of Figure 5 and achieve the same technical effect. To avoid repetition, it will not be repeated here.
  • Fig. 12 is a schematic diagram of the hardware structure of a terminal implementing an embodiment of the present application.
  • the terminal 1200 includes but is not limited to: a radio frequency unit 1201, a network module 1202, an audio output unit 1203, an input unit 1204, a sensor 1205, a display unit 1206, a user input unit 1207, an interface unit 1208, a memory 1209, and at least some of the components in the processor 1210.
  • the terminal 1200 may also include a power source (such as a battery) for supplying power to various components, and the power source may be logically connected to the processor 1210 through a power management system, so that the power management system can manage charging, discharging, and power consumption.
  • a power source such as a battery
  • the present invention may include more or fewer components than those shown in the figure, or some components may be combined, or the components may be arranged differently, which will not be described in detail here.
  • the input unit 1204 may include a graphics processing unit (GPU) 12041 and a microphone 12042, and the graphics processor 12041 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 1206 may include a display panel 12061, and the display panel 12061 may be configured in the form of a liquid crystal display, an organic light emitting diode, etc.
  • the user input unit 1207 includes a touch panel 12071 and at least one of other input devices 12072.
  • the touch panel 12071 is also called a touch screen.
  • the touch panel 12071 may include two parts: a touch detection device and a touch controller.
  • Other input devices 12072 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 1201 can transmit the data to the processor 1210 for processing; in addition, the RF unit 1201 can send uplink data to the network side device.
  • the RF unit 1201 includes but is not limited to an antenna, an amplifier, a transceiver, a coupler, a low noise amplifier, a duplexer, etc.
  • the memory 1209 can be used to store software programs or instructions and various data.
  • the memory 1209 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 1209 may include a volatile memory or a non-volatile memory, or the memory 1209 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 a flash memory.
  • the volatile memory may be a random access memory (RAM), a static random access memory (SRAM), a dynamic random access memory (DRAM), a synchronous dynamic random access memory (SDRAM), a double data rate synchronous dynamic random access memory (DDRSDRAM), an enhanced synchronous dynamic random access memory (ESDRAM), a synchronous link dynamic random access memory (SLDRAM) and a direct memory bus random access memory (DRRAM).
  • the memory 1209 in the embodiment of the present application includes but is not limited to these and any other suitable types of memory.
  • the processor 1210 may include one or more processing units; optionally, the processor 1210 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 1210.
  • the terminal provided in the embodiment of the present application can implement each process implemented by the method embodiment of Figure 4 and achieve the same technical effect. To avoid repetition, it will not be repeated here.
  • FIG. 13 is a structural diagram of a network side device applied in an embodiment of the present application.
  • the network side device 1300 includes: a processor 1301, a transceiver 1302, a memory 1303 and a bus interface, wherein the processor
  • the processor 1301 may be responsible for managing the bus architecture and general processing.
  • the memory 1303 may store data used by the processor 1301 when performing operations.
  • the network side device 1300 further includes: a program stored in the memory 1303 and executable on the processor 1301 , and when the program is executed by the processor 1301 , the steps in the method shown in FIG. 4 above are implemented.
  • the bus architecture may include any number of interconnected buses and bridges, specifically linking together various circuits of one or more processors represented by processor 1301 and memory represented by memory 1303.
  • the bus architecture may also link together various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein.
  • the bus interface provides an interface.
  • the transceiver 1302 may be a plurality of components, namely, a transmitter and a receiver, providing a unit for communicating with various other devices over a transmission medium.
  • an embodiment of the present application further provides a network side device 1300, including a processor 1401 and a memory 1402, and the memory 1402 stores programs or instructions that can be run on the processor 1401.
  • the communication device 1400 is a first node
  • the program or instruction is executed by the processor 1401 to implement the various steps of the method embodiment of Figure 4 above.
  • the communication device 1400 is a second node
  • the program or instruction is executed by the processor 1401 to implement the various steps of the method embodiment of Figure 5 above 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 readable storage medium, on which a program or instruction is stored.
  • a program or instruction is stored.
  • the method of Figure 4 or Figure 5 and the various processes of the above-mentioned embodiments 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 shown in Figure 4 or Figure 5 and the various method embodiments mentioned above, 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 shown in Figure 4 or Figure 5 and the various method embodiments described above, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.
  • An embodiment of the present application further provides a communication system, which includes a terminal and a network side device.
  • the terminal is used to execute the various processes as shown in Figure 4 and the various method embodiments described above
  • the network side device is used to execute the various processes as shown in Figure 5 and the various method embodiments described above, and can achieve the same technical effects. To avoid repetition, they are not repeated here.
  • the technical solution of the present application can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM/RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.
  • a storage medium such as ROM/RAM, a magnetic disk, or an optical disk
  • a terminal which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.

Landscapes

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

Abstract

本申请公开了一种更新AI模型参数的方法、装置及通信设备,该方法包括:第一节点向第二节点发送与AI模型参数相关的第一信息;其中,所述第一信息包括以下至少一项:所述AI模型参数的更新模式,所述AI模型参数的指示方式。

Description

更新AI模型参数的方法、装置及通信设备
相关申请的交叉引用
本申请主张在2022年09月26日在中国提交的中国专利申请No.202211177497.5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,具体涉及一种更新AI模型参数的方法、装置及通信设备。
背景技术
人工智能(Artificial Intelligence,AI)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。
相关技术中通过传输AI模型进行模型更新,然而这种更新方式效率较低以及网络信令开销较大。
发明内容
本申请实施例提供一种更新AI模型参数的方法、装置及通信设备,解决相关技术中AI模型的更新方式效率较低以及网络信令开销较大的问题。
第一方面,提供一种更新AI模型参数的方法,包括:
第一节点向第二节点发送与AI模型参数相关的第一信息;
其中,所述第一信息包括以下至少一项:所述AI模型参数的更新模式,所述AI模型参数的更新参数的指示方式。
第二方面,提供一种更新AI模型参数的方法,包括:
第二节点接收第一节点发送的与AI模型参数相关的第一信息;
所述第二节点根据所述第一信息更新所述AI模型参数;
其中,所述第一信息包括以下至少一项:所述AI模型参数的更新模式,所述AI模型参数的更新参数的指示方式。
第三方面,提供一种更新AI模型参数的装置,包括:
第一发送模块,用于向第二节点发送与AI模型参数相关的第一信息;
其中,所述第一信息包括以下至少一项:所述AI模型参数的更新模式,所述AI模型参数的更新参数的指示方式。
第四方面,提供一种更新AI模型参数的装置,包括:
第一接收模块,用于接收第一节点发送的与AI模型参数相关的第一信息;
更新模块,用于根据所述第一信息更新所述AI模型参数;
其中,所述第一信息包括以下至少一项:所述AI模型参数的更新模式,所述AI模型参数的更新参数的指示方式。
第五方面,提供了一种通信设备,包括:处理器,存储器及存储在所述存储器上并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二方面所述的方法的步骤。
第六方面,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或第二方面所述的方法的步骤。
第七方面,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,实现如第一方面或第二方面所述的法的步骤。
第八方面,提供了一种计算机程序/程序产品,所述计算机程序/程序产品被存储在非瞬态的存储介质中,所述程序/程序产品被至少一个处理器执行以实现如第一方面或第二方面所述的方法的步骤。
第九方面,提供一种通信系统,所述通信系统包括终端与网络侧设备,所述终端用于执行如第一方面所述的方法的步骤,所述网络侧设备用于执行如第二方面所述的方法的步骤。
在本申请实施例中,第一节点可以向第二节点发送所述AI模型参数的更新模式和/或AI模型参数的更新参数的指示方式,第二节点不需要对AI模型进行编译或重编译,第二节点根据接收到的AI模型参数的更新模式和/或AI模型参数的更新参数的指示方式更新已有AI模型参数,能够有效提升无线通信系统中传输AI模型的效率,以及减少网络信令开销。
附图说明
图1为神经网络的示意图;
图2为神经元的示意图;
图3为本申请实施例的无线通信系统的架构示意图;
图4为本申请实施例的更新AI模型参数的方法的流程图之一;
图5为本申请实施例的更新AI模型参数的方法的流程图之二;
图6为本申请实施例的AI模型参数的指示方式的示意图之一;
图7为本申请实施例的AI模型参数的指示方式的示意图之二;
图8为本申请实施例的AI模型参数的指示方式的示意图之三;
图9为本申请实施例的AI模型参数的指示方式的示意图之四;
图10为本申请实施例的更新AI模型参数的装置的结构图之一;
图11为本申请实施例的更新AI模型参数的装置的结构图之二;
图12为本申请实施例的终端的示意图;
图13为本申请实施例的网络侧设备的示意图;
图14为本申请实施例的通信设备的示意图。
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)系统,还可用于其他无线通信系统,诸如码分多址(Code Division Multiple Access,CDMA)、时分多址(Time Division Multiple Access,TDMA)、频分多址(Frequency Division Multiple Access,FDMA)、正交频分多址(Orthogonal Frequency Division Multiple Access,OFDMA)、单载波频分多址(Single-carrier Frequency Division Multiple Access,SC-FDMA)和其他系统。本申请实施例中的术语“系统”和“网络”常被可互换地使用,所描述的技术既可用于以上提及的系统和无线电技术,也可用于其他系统和无线电技术。以下描述出于示例目的描述了新空口(New Radio,NR)系统,并且在以下大部分描述中使用NR术语,但是这些技术也可应用于NR系统应用以外的应用,如第6代(6th Generation,6G)通信系统。
为了便于理解本申请的实施方式,下面先介绍以下技术点。
1、关于神经网络的介绍
本申请以神经网络为例进行说明,但是并不限定AI模块的具体类型,神经网络的结构如图1所示。
其中,神经网络由神经元组成,神经元的示意图如图2所示。其中a1,a2,…aK为输入,w为权值(乘性系数),b为偏置(加性系数),σ(.)为激活函数,z=a1*w1+…+ak*wk+…+aK*wK+b。常见的激活函数包括Sigmoid函数、tanh函数、修正线性单元(Rectified Linear Unit,ReLU)等等。
神经网络的参数可以通过优化算法进行优化。优化算法就是一种能够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,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。如果找到合适的W,b使上述的损失函数的值达到最小,损失值越小,则说明模型越接近于真实情况。
目前常见的优化算法,基本都是基于误差反向传播(error Back Propagation,BP)算法。BP算法的基本思想是,学习过程由信号的正向传播与误差的反向传播两个过程组成。正向传播时,输入样本从输入层传入,经各隐藏层逐层处理后,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藏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。根据相关技术,借助AI提升第五代移动通信技术(5th Generation,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。
常见的优化算法有梯度下降(Gradient Descent)、随机梯度下降(Stochastic Gradient Descent,SGD)、小批量梯度下降(mini-batch gradient descent)、动量法(Momentum)、内斯特洛夫(Nesterov)加速算法、自适应梯度下降(ADAptive GRADient descent,Adagrad)、Adadelta、均方根误差降速(Root Mean Square prop,RMSprop)、自适应动量估计(Adaptive Moment Estimation,Adam)等。
这些优化算法在误差反向传播时,都是根据损失函数得到的误差/损失,对当前神经元求导数/偏导,加上学习速率、之前的梯度/导数/偏导等影响,得到梯度,将梯度传给上一层。图3示出本申请实施例可应用的一种无线通信系统的框图。无线通信系统包括终端31和网络侧设备32。
其中,终端31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。除了上述终端设备,本申请涉及的终端也可以是终端内的芯片,例如调制解调器(Modem)芯片,系统级芯片(System on Chip,SoC)。需要说明的是,在本申请实施例并不限定终端31的具体类型。
网络侧设备32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(eNB)、接入点、基收发机站(Base Transceiver Station,BTS)、无线电基站、无线电收发机、基本服务集(Basic Service Set,BSS)、扩展服务集(Extended Service Set,ESS)、家用B节点、家用演进型B节点、发送接收点(Transmitting Receiving Point,TRP)或所述领域中其他某个合适的术语,只要达到相同的技术效果,所述基站不限于特定技术词汇,需要说明的是,在本申请实施例中仅以NR系统中的基站为例进行介绍,并不限定基站的具体类型。
核心网设备可以包含但不限于如下至少一项:核心网节点、核心网功能、移动管理实体(Mobility Management Entity,MME)、接入和移动管理功能(Access and Mobility Management Function,AMF)、会话管理功能(Session Management Function,SMF)、用户平面功能(User Plane Function,UPF)、策略控制功能(Policy Control Function,PCF)、策略与计费规则功能单元(Policy and Charging Rules Function,PCRF)、边缘应用服务发现功能(Edge Application Server Discovery Function,EASDF)、统一数据管理(Unified Data Management,UDM),统一数据仓储(Unified Data Repository,UDR)、归属用户服务器(Home Subscriber Server,HSS)、集中式网络配置(Centralized network configuration,CNC)、网络存储功能(Network Repository Function,NRF),网络开放功能(Network Exposure Function,NEF)、本地NEF(Local NEF,或L-NEF)、绑定支持功能(Binding Support Function,BSF)、应用功能(Application Function,AF)、非3GPP互通功能(Non-3GPP InterWorking Function,N3IWF)等。需要说明的是,在本申请实施例中仅以NR系统中的核心网设备为例进行介绍,并不限定核心网设备的具体类型。
下面结合附图,通过一些实施例及其应用场景对本申请实施例提供的更新AI模型参数的方法、装置、通信设备及可读存储介质进行详细地说明。
参见图4,本申请实施例提供一种更新AI模型参数的方法,应用于第一节点,该第一节点可以是网络侧设备或终端,具体步骤包括:步骤401。
步骤401:第一节点向第二节点发送与AI模型参数相关的第一信息;
其中,所述第一信息包括以下至少一项:所述AI模型参数的更新模式,所述AI模型参数的指示方式。
在本实施例中,所述第一信息用于第二节点更新AI模型参数,第一信息中的AI模型参数的更新模式用于表示采用哪种方式更新AI模型参数,第一信息中的AI模型参数的指 示方式用于指示更新参数的位置或者用于指示AI模型参数的发送顺序,比如第一节点可以向第二节点指示所有AI模型参数的发送顺序,第二节点根据情况对接收到的所有AI模型参数中的部分参数进行AI模型更新,或者第一节点可以向第二节点指示部分AI模型参数的发送顺序,第二节点根据接收到的部分AI模型参数进行AI模型更新,其中的部分AI模型参数也可以称为AI模型更新参数。
在本实施例中,第一节点可以向第二节点发送所述AI模型参数的更新模式和/或AI模型参数的指示方式,第二节点不需要对AI模型进行编译或重编译,第二节点根据接收到的AI模型参数的更新模式和/或AI模型参数的指示方式更新已有AI模型参数,能够有效提升无线通信系统中传输AI模型的效率,以及减少网络信令开销。
在本申请的一种实施例中,所述第一节点为第一网络侧设备或第一终端,所述第二节点为第二网络侧设备或第二终端,比如,第一节点为第一网络侧设备,第二节点为第二终端,或者,第一节点为第一终端,第二节点为第二网络侧设备,或者,第一节点为第一网络侧设备,第二节点为第二网络侧设备,或者第一节点为第一终端,第二节点为第二终端。
在本申请的一种实施例中,所述更新模式包括以下之一:
(1)第一模式;
可选的,所述第一模式包括:更新第一参数子集中的所有参数,所述第一参数子集是所述AI模型参数的子集,所述第一参数子集是预配置或者协议约定的或者网络侧指示的;
(2)第二模式;
可选的,所述第二模式包括:更新第二参数子集中的所有参数,所述第二参数子集是所述第一参数子集的子集;
(3)第三模式;
可选的,所述第三模式包括:更新所述AI模型的所有参数;
(4)第四模式;
可选的,所述第四模式包括:更新第三参数子集,所述第三参数子集是所述AI模型参数的子集。
在本申请的一种实施例中,所述指示方式包括以下之一:
(1)第一信息,所述第一信息用于指示更新参数在所述AI模型中的位置;
(2)第二信息,所述第二信息用于指示所述AI模型中更新参数的发送顺序。
示例性的,本申请实施例适用的一种神经网络的结构如图1所示,一层对应一列,即输入层为第一列,输出层为最后一列,隐藏层为中间列,当然本申请实施例也可以适用其他结构的神经网络。
在一种实施方式中,可以基于图1所示的神经网络的结构“指示更新参数在所述AI模型中的位置”,以及“指示所述AI模型中更新参数的发送顺序”。
在另一种实施方式中,神经网络的结构可以是根据图1旋转得到的神经网络的另一种结构,这样基于该神经网络的另一种结构“指示更新参数在所述AI模型中的位置”,以及 “指示所述AI模型中更新参数的发送顺序”,指示方式应用在图1所示的神经网络的结构,与应用在神经网络的其他结构相似,在此不再赘述。
比如,神经网络的另一种结构可以是相对于图1的示意向右旋转90度,即一层对应一行,输入层为第一行,输出层为最后一行,隐藏层为中间行;又比如,神经网络的另一种结构可以是相对于图1的示意的神经网络结构向左旋转90度,同样一层对应一行,输入层为最后一行,输出层为第一行,隐藏层为中间行,可以理解的是神经网络的结构并不限于以上三种形式。
在本申请的一种实施例中,所述AI模型为神经网络,所述神经网络包括输入层、一个或多个隐藏层和输出层,所述输入层、一个或多个隐藏层、输出层,以及各层的神经元组成多行多列的结构,所述神经网络中神经元的参数的发送顺序是根据所述神经元在所述神经网络中的位置确定的。
在本申请的一种实施例中,所述神经网络中神经元的参数的发送顺序,包括以下之一:
(1)第一列所有神经元的参数,第二列所有神经元的参数,以此类推,直至最后一列所有神经元的参数;
在一种实施方式中,所述第一列对应所述输入层,所述最后一列对应所述输出层;或者,所述第一列对应所述输出层,所述最后一列对应所述输入层。
(2)第一行所有神经元的参数,第二行所有神经元的参数,以此类推,直至最后一行所有神经元的参数。
在一种实施方式中,所述第一行对应所述输入层,所述最后一行对应所述输出层;或者,所述第一行对应所述输出层,所述最后一行对应所述输入层;
在一种实施方式中,所述第一行是所述神经网络中最上方一行,所述最后一行是所述神经网络中最下方一行;或者,所述第一行是所述神经网络中最下方一行,所述最后一行是所述神经网络中最上方一行。
在本申请的一种实施例中,在所述神经网络中神经元的参数的发送顺序包括:第一列所有神经元的参数,第二列所有神经元的参数,以此类推,直至最后一列所有神经元的参数的情况下,所述神经网络中神经元的参数的发送顺序还包括以下之一:
(1)每一列中神经元的参数按照从上往下的神经元的顺序发送;
(2)每一列中神经元的参数按照从下往上的神经元的顺序发送;
(3)每一列中神经元的参数的发送顺序相同,如图6所示;
(4)相邻列的神经元的参数的发送顺序相反,如图7所示。
在本申请的一种实施例中,在所述神经网络中神经元的参数的发送顺序包括第一行所有神经元的参数,第二行所有神经元的参数,以此类推,直至最后一行所有神经元的参数的情况下,所述神经网络中神经元的参数的发送顺序还包括以下之一:
(1)每一行中神经元的参数按照从左往右的神经元的顺序发送;
(2)每一行中神经元的参数按照从右往左的神经元的顺序发送;
(3)每一行中神经元的参数的发送顺序相同,如图8所示;
(4)相邻行的神经元的参数的发送顺序相反,如图9所示。
在本申请的一种实施例中,当每层的神经元数目不是完全相同时,通过以下至少一项发送神经元的参数:
(1)每一行都按最多神经元数量的行发送所述神经元的参数,缺少神经元的层补预设值(比如,0)。
比如,第一层50个神经元,第二层40个神经元,第三层60个神经元,第四层30个神经元,如第50行,第二层、第四层没有神经元,则发送时第50行的第二层、第四层补0。
(2)每一行都按实际神经元数量发送所述神经元的参数。
比如,第一层50个神经元,第二层40个神经元,第三层60个神经元,第四层30个神经元,如第50行,第二层、第四层没有神经元,则发送时跳过第二层和第四层,第50行发送第一层和第三层的神经元的参数。
在本申请的一种实施例中,所述AI模型参数的发送顺序是预配置的或协议约定的或者网络侧指示的。
在本申请的一种实施例中,所述AI模型参数包括乘性系数和加性系数,所述乘性系数的发送优先级高于加性系数的发送优先级;或者,所述加性系数的发送优先级高于乘性系数的发送优先级。
在本申请的一种实施例中,所述AI模型的量化等级和/或压缩方式是预配置的或者协议约定的或者网络侧指示的。
在本申请的一种实施例中,更新后的AI模型的模型标识(Identity,ID)与更新前的AI模型的模型ID相同,不保存所述更新前的AI模型;
或者,
更新后的AI模型的模型ID与更新前的AI模型的模型ID不同,保存所述更新前的AI模型,且所述更新前的AI模型的模型ID不变。
在本申请的一种实施方式中,所述AI模型包括第一功能模块,所述第一功能模块用于以下至少一项:
(1)信号处理,包括但不限于以下至少一项:信号检测、滤波、均衡等,其中信号包括但不限于以下至少一项:解调参考信号(Demodulation Reference Signal,DMRS)、探测参考信号(Sounding Reference Signal,SRS)、同步信号块(Synchronization Signal and PBCH block,SSB)、跟踪参考信号(Tracking Reference Signal,TRS)、相位跟踪参考信号(Phase-tracking reference signals,PTRS)、信道状态信息参考信号(Channel-State-Information Reference Signal,CSI-RS)等;
(2)信道信号传输、信道信号接收、信道解调或信道信号发送,其中信道包括但不限于以下至少一项:物理下行控制信道(Physical Downlink Control Channel,PDCCH)、 物理下行共享信道(Physical Downlink Shared Channel,PDSCH)、物理上行控制信道(Physical Uplink Control Channel,PUCCH)、物理上行共享信道(Physical Uplink Shared Channel,PUSCH)、物理随机接入信道(Physical Random Access Channel,PRACH)、物理广播信道(Physical Broadcast Channel,PBCH);
(3)获取处理信道状态信息;
比如,信道状态信息反馈,包括但不限于以下至少一项:信道相关信息、信道矩阵相关信息、信道特征信息、信道矩阵特征信息、预编码矩阵指示(Precoding matrix indicator,PMI)、秩指示(Rank indicator,RI)、CSI-RS资源指示(CSI-RS Resource Indicator,CRI)、信道质量指示(Channel quality indicator,CQI)、层指示(Layer Indicator,LI)等。
又比如,频分双工(Frequency Division Duplex,FDD)上下行部分互易性。对于FDD系统,根据部分互异性,基站根据上行信道获取角度和时延信息,可以通过CSI-RS预编码或者直接指示的方法,将角度信息和时延信息通知终端,终端根据基站的指示上报或者在基站的指示范围内选择并上报,从而减少终端的计算量和CSI上报的开销。
(4)波束管理,包括但不限于以下至少一项:波束测量、波束上报、波束预测、波束失败检测、波束失败恢复、波束失败恢复中的新波束指示;
(5)信道预测,包括但不限于以下至少一项:信道状态信息的预测、波束预测;
(6)干扰抑制,包括但不限于以下至少一项:小区内干扰、小区间干扰、带外干扰、交调干扰等;
(7)定位,比如通过参考信号(例如SRS),估计出终端的具体位置(包括水平位置和/或垂直位置)或未来可能的轨迹,或估计出终端的辅助位置估计或轨迹估计的信息;
(8)预测或管理高层业务和/或高层参数,包括但不限于以下至少一项:吞吐量、所需数据包大小、业务需求、移动速度、噪声信息等;
(9)解析控制信令,包括但不限于以下至少一项:功率控制的相关信令,波束管理的相关信令。
在本申请实施例中,能够有效提升无线通信系统中传输AI模型的效率,以及减少网络信令开销。
参见图5,本申请实施例提供一种更新AI模型参数的方法,应用于第二节点,该第二节点为网络侧设备或终端,该方法包括:步骤501和步骤502。
步骤501:第二节点接收第一节点发送的与AI模型参数相关的第一信息;
步骤502:第二节点根据所述第一信息更新所述AI模型参数;
其中,所述第一信息包括以下至少一项:所述AI模型参数的更新模式,所述AI模型参数的指示方式。
在本申请的一种实施例中,所述第一节点为第一网络侧设备或第一终端,所述第二节点为第二网络侧设备或第二终端,比如,第一节点为第一网络侧设备,第二节点为第二终 端,或者,第一节点为第一终端,第二节点为第二网络侧设备,或者,第一节点为第一网络侧设备,第二节点为第二网络侧设备,或者第一节点为第一终端,第二节点为第二终端。
在本申请的一种实施例中,所述更新模式包括以下之一:
(1)第一模式;
可选的,所述第一模式包括:更新第一参数子集中的所有参数,所述第一参数子集是所述AI模型参数的子集,所述第一参数子集是预配置或者协议约定的或者网络侧指示的;
(2)第二模式;
可选的,所述第二模式包括:更新第二参数子集中的所有参数,所述第二参数子集是所述第一参数子集的子集;
(3)第三模式;
可选的,所述第三模式包括:更新所述AI模型的所有参数;
(4)第四模式;
可选的,所述第四模式包括:更新第三参数子集,所述第三参数子集是所述AI模型参数的子集。
在本申请的一种实施例中,所述指示方式包括以下之一:
(1)第一信息,所述第一信息用于指示所述AI模型参数在AI模型中的位置;
(2)第二信息,所述第二信息用于指示所述AI模型中更新参数的发送顺序。
在本申请的一种实施例中,所述AI模型为神经网络,所述神经网络包括输入层、一个或多个隐藏层和输出层,所述输入层、一个或多个隐藏层、输出层,以及各层的神经元组成多行多列的结构,所述神经网络中各个神经元的参数的发送顺序是根据所述神经元在所述神经网络中的位置确定的。
在本申请的一种实施例中,所述神经网络中各个神经元的参数的发送顺序,包括:
第一列所有神经元的参数,第二列所有神经元的参数,以此类推,直至最后一列所有神经元的参数;
或者,
第一行所有神经元的参数,第二行所有神经元的参数,以此类推,直至最后一行所有神经元的参数。
在本申请的一种实施例中,所述第一列对应所述输入层,所述最后一列对应所述输出层;或者,所述第一列对应所述输出层,所述最后一列对应所述输入层;
或者,
所述第一行对应所述输入层,所述最后一行对应所述输出层;或者,所述第一行对应所述输出层,所述最后一行对应所述输入层。
在本申请的一种实施例中,每一列中神经元的参数按照从上往下的神经元的顺序发送,或者每一列中神经元的参数按照从下往上的神经元的顺序发送,或者每一列中神经元的参数的发送顺序相同,或者相邻列的神经元的参数的发送顺序相反。
在本申请的一种实施例中,所述第一行是所述神经网络中最上方一行,所述最后一行是所述神经网络中最下方一行;或者,所述第一行是所述神经网络中最下方一行,所述最后一行是所述神经网络中最上方一行。
在本申请的一种实施例中,每一行中神经元的参数按照从左往右的神经元的顺序发送,或者每一行中神经元的参数按照从右往左的神经元的顺序发送,或者每一行中神经元的参数的发送顺序相同,或者相邻行的神经元的参数的发送顺序相反。
在本申请的一种实施例中,在每层的神经元数目不是完全相同的情况下,每一行都按最多神经元数量的行发送所述神经元的参数,缺少神经元的层补预设值,或者,每一行都按实际神经元数量发送所述神经元的参数。
在本申请的一种实施例中,所述AI模型参数的发送顺序是预配置的或协议约定的或者网络侧指示的。
在本申请的一种实施例中,所述AI模型参数包括乘性系数和加性系数,所述AI模型中乘性系数的发送优先级高于加性系数的发送优先级;或者,所述AI模型中加性系数的发送优先级高于乘性系数的发送优先级。
在本申请的一种实施例中,所述AI模型的量化等级和/或压缩方式是预配置的或者协议约定的或者网络侧指示的。
在本申请的一种实施例中,更新后的AI模型的模型ID与更新前的AI模型的模型ID相同,不保存所述更新前的AI模型;
或者,
更新后的AI模型的模型ID与更新前的AI模型的模型ID不同,保存所述更新前的AI模型,且所述更新前的AI模型的模型ID不变。
在本申请的一种实施方式中,所述AI模型包括第一功能模块,所述第一功能模块用于以下至少一项:
(1)信号处理,包括但不限于以下至少一项:信号检测、滤波、均衡等,其中信号包括但不限于以下至少一项:DMRS、SRS、SSB、TRS、PTRS、CSI-RS等;
(2)信道信号传输、信道信号接收、信道解调或信道信号发送,其中信道包括但不限于以下至少一项:PDCCH、PDSCH、PUCCH、PUSCH、PRACH、PBCH;
(3)获取处理信道状态信息;
比如,信道状态信息反馈,包括但不限于以下至少一项:信道相关信息、信道矩阵相关信息、信道特征信息、信道矩阵特征信息、PMI、RI、CRI、CQI、LI等。
又比如,FDD上下行部分互易性。对于FDD系统,根据部分互异性,基站根据上行信道获取角度和时延信息,可以通过CSI-RS预编码或者直接指示的方法,将角度信息和时延信息通知终端,终端根据基站的指示上报或者在基站的指示范围内选择并上报,从而减少终端的计算量和CSI上报的开销。
(4)波束管理,包括但不限于以下至少一项:波束测量、波束上报、波束预测、波 束失败检测、波束失败恢复、波束失败恢复中的新波束指示;
(5)信道预测,包括但不限于以下至少一项:信道状态信息的预测、波束预测;
(6)干扰抑制,包括但不限于以下至少一项:小区内干扰、小区间干扰、带外干扰、交调干扰等;
(7)定位,比如通过参考信号(例如SRS),估计出终端的具体位置(包括水平位置和/或垂直位置)或未来可能的轨迹,或估计出终端的辅助位置估计或轨迹估计的信息;
(8)预测或管理高层业务和/或高层参数,包括但不限于以下至少一项:吞吐量、所需数据包大小、业务需求、移动速度、噪声信息等;
(9)解析控制信令,包括但不限于以下至少一项:功率控制的相关信令,波束管理的相关信令。
在本申请实施例中,能够有效提升无线通信系统中传输AI模型的效率,以及减少网络信令开销。
参见图10,本申请实施例提供一种更新AI模型参数的装置,应用于第一节点,该装置1000包括:
第一发送模块1001,用于向第二节点发送与AI模型参数相关的第一信息;
其中,所述第一信息包括以下至少一项:所述AI模型参数的更新模式,所述AI模型参数的指示方式。
在本申请的一种实施方式中,所述更新模式包括以下之一:第一模式、第二模式、第三模式、第四模式;
其中,所述第一模式包括:更新第一参数子集中的所有参数,所述第一参数子集是所述AI模型参数的子集,所述第一参数子集是预配置或者协议约定的或者网络侧指示的;
所述第二模式包括:更新第二参数子集中的所有参数,所述第二参数子集是所述第一参数子集的子集;
所述第三模式包括:更新所述AI模型的所有参数;
所述第四模式包括:更新第三参数子集,所述第三参数子集是所述AI模型参数的子集。
在本申请的一种实施方式中,所述指示方式包括以下之一:
第一信息,所述第一信息用于指示所述AI模型参数在所述AI模型中的位置;
第二信息,所述第二信息用于指示所述AI模型参数的发送顺序。
在本申请的一种实施方式中,所述AI模型为神经网络,所述神经网络包括输入层、一个或多个隐藏层和输出层,所述输入层、一个或多个隐藏层、输出层,以及各层的神经元组成多行多列的结构,所述神经网络中各个神经元的参数的发送顺序是根据所述神经元在所述神经网络中的位置确定的。
在本申请的一种实施方式中,所述神经网络中各个神经元的参数的发送顺序,包括:
第一列所有神经元的参数,第二列所有神经元的参数,以此类推,直至最后一列所有 神经元的参数;或,
第一行所有神经元的参数,第二行所有神经元的参数,以此类推,直至最后一行所有神经元的参数。
在本申请的一种实施方式中,所述第一列对应所述输入层,所述最后一列对应所述输出层,或者,所述第一列对应所述输出层,所述最后一列对应所述输入层;
或者,
所述第一行对应所述输入层,所述最后一行对应所述输出层,或者,所述第一行对应所述输出层,所述最后一行对应所述输入层。
在本申请的一种实施方式中,在所述神经网络中神经元的参数的发送顺序包括第一列所有神经元的参数,第二列所有神经元的参数,以此类推,直至最后一列所有神经元的参数的情况下,所述神经网络中神经元的参数的发送顺序还包括以下之一:
每一列中神经元的参数按照从上往下的神经元的顺序发送;
每一列中神经元的参数按照从下往上的神经元的顺序发送;
每一列中神经元的参数的发送顺序相同;
相邻列的神经元的参数的发送顺序相反。
在本申请的一种实施方式中,所述第一行是所述神经网络中最上方一行,所述最后一行是所述神经网络中最下方一行;或者,所述第一行是所述神经网络中最下方一行,所述最后一行是所述神经网络中最上方一行。
在本申请的一种实施方式中,在所述神经网络中神经元的参数的发送顺序包括第一行所有神经元的参数,第二行所有神经元的参数,以此类推,直至最后一行所有神经元的参数的情况下,所述神经网络中神经元的参数的发送顺序还包括以下之一:
每一行中神经元的参数按照从左往右的神经元的顺序发送;
每一行中神经元的参数按照从右往左的神经元的顺序发送;
每一行中神经元的参数的发送顺序相同;
相邻行的神经元的参数的发送顺序相反。
在本申请的一种实施方式中,所述AI模型参数的发送顺序是预配置的或协议约定的或者网络侧指示的。
在本申请的一种实施方式中,所述AI模型参数包括乘性系数和加性系数,所述乘性系数的发送优先级高于加性系数的发送优先级;或者,所述加性系数的发送优先级高于乘性系数的发送优先级。
在本申请的一种实施方式中,所述AI模型的量化等级和/或压缩方式是预配置的或者协议约定的或者网络侧指示的。
在本申请的一种实施方式中,更新后的AI模型的模型ID与更新前的AI模型的模型ID相同,不保存所述更新前的AI模型;
或者,
更新后的AI模型的模型ID与更新前的AI模型的模型ID不同,保存所述更新前的AI模型,且所述更新前的AI模型的模型ID不变。
在本申请的一种实施方式中,所述AI模型包括第一功能模块,所述第一功能模块用于以下一项或多项:
信号处理;
信道传输、信道接收、信道解调或信道发送;
获取信道状态信息;
波束管理;
信道预测;
干扰抑制;
定位;
预测或管理高层业务和/或高层参数的预测;
解析控制信令。
在本申请的一种实施方式中,所述第一节点为第一网络侧设备或第一终端,所述第二节点为第二网络侧设备或第二终端。
本申请实施例提供的装置能够实现图4的方法实施例实现的各个过程,并达到相同的技术效果,为避免重复,这里不再赘述。
参见图11,本申请实施例提供一种更新AI模型参数的装置,应用于第二节点,该装置1100包括:
第一接收模块1101,用于接收第一节点发送的与AI模型参数相关的第一信息;
更新模块1102,用于根据所述第一信息更新所述AI模型参数;
其中,所述第一信息包括以下至少一项:所述AI模型参数的更新模式,所述AI模型参数的指示方式。
在本申请的一种实施方式中,所述更新模式包括以下之一:第一模式、第二模式、第三模式、第四模式;
其中,所述第一模式包括:更新第一参数子集中的所有参数,所述第一参数子集是所述AI模型参数的子集,所述第一参数子集是预配置或者协议约定的或者网络侧指示的;
所述第二模式包括:更新第二参数子集中的所有参数,所述第二参数子集是所述第一参数子集的子集,所述第一参数子集是所述AI模型参数的子集,所述第一参数子集是预先指示或协议约定的;
所述第三模式包括:更新所述AI模型的所有参数;
所述第四模式包括:更新第三参数子集,所述第三参数子集是所述AI模型参数的子集。
在本申请的一种实施方式中,所述指示方式包括以下之一:
第一信息,所述第一信息用于指示所述AI模型参数在所述AI模型中的位置;
第二信息,所述第二信息用于指示所述AI模型参数的发送顺序。
在本申请的一种实施方式中,所述AI模型为神经网络,所述神经网络包括输入层、一个或多个隐藏层和输出层,所述输入层、一个或多个隐藏层、输出层,以及各层的神经元组成多行多列的结构,所述神经网络中各个神经元的参数的发送顺序是根据所述神经元在所述神经网络中的位置确定的。
在本申请的一种实施方式中,所述神经网络中各个神经元的参数的发送顺序,包括:
第一列所有神经元的参数,第二列所有神经元的参数,以此类推,直至最后一列所有神经元的参数;
或者,
第一行所有神经元的参数,第二行所有神经元的参数,以此类推,直至最后一行所有神经元的参数。
在本申请的一种实施方式中,所述第一列对应所述输入层,所述最后一列对应所述输出层,或者,所述第一列对应所述输出层,所述最后一列对应所述输入层;
或者,
所述第一行对应所述输入层,所述最后一行对应所述输出层,或者,所述第一行对应所述输出层,所述最后一行对应所述输入层。
在本申请的一种实施方式中,在所述神经网络中神经元的参数的发送顺序包括第一列所有神经元的参数,第二列所有神经元的参数,以此类推,直至最后一列所有神经元的参数的情况下,所述神经网络中神经元的参数的发送顺序还包括以下之一:
每一列中神经元的参数按照从上往下的神经元的顺序发送;
每一列中神经元的参数按照从下往上的神经元的顺序发送;
每一列中神经元的参数的发送顺序相同;
相邻列的神经元的参数的发送顺序相反。
在本申请的一种实施方式中,所述第一行是所述神经网络中最上方一行,所述最后一行是所述神经网络中最下方一行;或者,所述第一行是所述神经网络中最下方一行,所述最后一行是所述神经网络中最上方一行。
在本申请的一种实施方式中,在所述神经网络中神经元的参数的发送顺序包括第一行所有神经元的参数,第二行所有神经元的参数,以此类推,直至最后一行所有神经元的参数的情况下,所述神经网络中神经元的参数的发送顺序还包括以下之一:
每一行中神经元的参数按照从左往右的神经元的顺序发送;
每一行中神经元的参数按照从右往左的神经元的顺序发送;
每一行中神经元的参数的发送顺序相同;
相邻行的神经元的参数的发送顺序相反。
在本申请的一种实施方式中,在每层的神经元数目不是完全相同的情况下,每一行都按最多神经元数量的行发送所述神经元的参数,缺少神经元的层补预设值,或者,每一行 都按实际神经元数量发送所述神经元的参数。
在本申请的一种实施方式中,所述AI模型参数的发送顺序是预配置的或协议约定的或者网络侧指示的。
在本申请的一种实施方式中,所述AI模型参数包括乘性系数和加性系数,乘性系数的发送优先级高于加性系数的发送优先级;或者,加性系数的发送优先级高于乘性系数的发送优先级。
在本申请的一种实施方式中,所述AI模型的量化等级和/或压缩方式是预配置的或者协议约定的或者网络侧指示的。
在本申请的一种实施方式中,更新后的AI模型的模型ID与更新前的AI模型的模型ID相同,不保存所述更新前的AI模型;
或者,
更新后的AI模型的模型ID与更新前的AI模型的模型ID不同,保存所述更新前的AI模型,且所述更新前的AI模型的模型ID不变。
在本申请的一种实施方式中,所述AI模型包括第一功能模块,所述第一功能模块用于以下一项或多项:
信号处理;
信道传输、信道接收、信道解调或信道发送;
获取信道状态信息;
波束管理;
信道预测;
干扰抑制;
定位;
预测或管理高层业务和/或高层参数的预测;
解析控制信令。
在本申请的一种实施方式中,所述第二节点为第二网络侧设备或第二终端,所述第一节点为第一网络侧设备或第一终端。
本申请实施例提供的装置能够实现图5的方法实施例实现的各个过程,并达到相同的技术效果,为避免重复,这里不再赘述。
图12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1200包括但不限于:射频单元1201、网络模块1202、音频输出单元1203、输入单元1204、传感器1205、显示单元1206、用户输入单元1207、接口单元1208、存储器1209以及处理器1210等中的至少部分部件。
本领域技术人员可以理解,终端1200还可以包括给各个部件供电的电源(比如电池),电源可以通过电源管理系统与处理器1210逻辑相连,从而通过电源管理系统实现管理充电、放电、以及功耗管理等功能。图12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1204可以包括图形处理单元(Graphics Processing Unit,GPU)12041和麦克风12042,图形处理器12041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1206可包括显示面板12061,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12061。用户输入单元1207包括触控面板12071以及其他输入设备12072中的至少一种。触控面板12071,也称为触摸屏。触控面板12071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12072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1201接收来自网络侧设备的下行数据后,可以传输给处理器1210进行处理;另外,射频单元1201可以向网络侧设备发送上行数据。通常,射频单元1201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1209可用于存储软件程序或指令以及各种数据。存储器1209可主要包括存储程序或指令的第一存储区和存储数据的第二存储区,其中,第一存储区可存储操作系统、至少一个功能所需的应用程序或指令(比如声音播放功能、图像播放功能等)等。此外,存储器1209可以包括易失性存储器或非易失性存储器,或者,存储器1209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1209包括但不限于这些和任意其它适合类型的存储器。
处理器1210可包括一个或多个处理单元;可选的,处理器1210集成应用处理器和调制解调处理器,其中,应用处理器主要处理涉及操作系统、用户界面和应用程序等的操作,调制解调处理器主要处理无线通信信号,如基带处理器。可以理解的是,上述调制解调处理器也可以不集成到处理器1210中。
本申请实施例提供的终端能够实现图4的方法实施例实现的各个过程,并达到相同的技术效果,为避免重复,这里不再赘述。
请参阅图13,图13是本申请实施例应用的网络侧设备的结构图,如图13所示,网络侧设备1300包括:处理器1301、收发机1302、存储器1303和总线接口,其中,处理 器1301可以负责管理总线架构和通常的处理。存储器1303可以存储处理器1301在执行操作时所使用的数据。
在本申请的一个实施例中,网络侧设备1300还包括:存储在存储器1303并可在处理器1301上运行的程序,程序被处理器1301执行时实现以上图4所示方法中的步骤。
在图13中,总线架构可以包括任意数量的互联的总线和桥,具体由处理器1301代表的一个或多个处理器和存储器1303代表的存储器的各种电路链接在一起。总线架构还可以将诸如外围设备、稳压器和功率管理电路等之类的各种其他电路链接在一起,这些都是本领域所公知的,因此,本文不再对其进行进一步描述。总线接口提供接口。收发机1302可以是多个元件,即包括发送机和接收机,提供用于在传输介质上与各种其他装置通信的单元。
可选的,如图14所示,本申请实施例还提供一种网络侧设备1300,包括处理器1401和存储器1402,存储器1402上存储有可在所述处理器1401上运行的程序或指令,例如,该通信设备1400为第一节点时,该程序或指令被处理器1401执行时实现上述图4方法实施例的各个步骤,该通信设备1400为第二节点时,该程序或指令被处理器1401执行时实现上述图5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图4或图5方法及上述各个实施例的各个过程,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,实现图4或图5所示及上述各个方法实施例的各个过程,且能达到相同的技术效果,为避免重复,这里不再赘述。
应理解,本申请实施例提到的芯片还可以称为系统级芯片,系统芯片,芯片系统或片上系统芯片等。
本申请实施例另提供了一种计算机程序/程序产品,所述计算机程序/程序产品被存储在存储介质中,所述计算机程序/程序产品被至少一个处理器执行以实现图4或图5所示及上述各个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。
本申请实施例另提供一种通信系统,所述通信系统包括终端与网络侧设备,所述终端用于执行如图4及上述各个方法实施例的各个过程,所述网络侧设备用于执行如图5及上述各个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。
需要说明的是,在本文中,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、物品或者装置不仅包括那些要素, 而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、物品或者装置所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、方法、物品或者装置中还存在另外的相同要素。此外,需要指出的是,本申请实施方式中的方法和装置的范围不限按示出或讨论的顺序来执行功能,还可包括根据所涉及的功能按基本同时的方式或按相反的顺序来执行功能,例如,可以按不同于所描述的次序来执行所描述的方法,并且还可以添加、省去、或组合各种步骤。另外,参照某些示例所描述的特征可在其他示例中被组合。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件,但很多情况下前者是更佳的实施方式。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以计算机软件产品的形式体现出来,该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中,包括若干指令用以使得一台终端(可以是手机,计算机,服务器,空调器,或者网络设备等)执行本申请各个实施例所述的方法。
上面结合附图对本申请的实施例进行了描述,但是本申请并不局限于上述的具体实施方式,上述的具体实施方式仅仅是示意性的,而不是限制性的,本领域的普通技术人员在本申请的启示下,在不脱离本申请宗旨和权利要求所保护的范围情况下,还可做出很多形式,均属于本申请的保护之内。

Claims (36)

  1. 一种更新AI模型参数的方法,包括:
    第一节点向第二节点发送与AI模型参数相关的第一信息;
    其中,所述第一信息包括以下至少一项:所述AI模型参数的更新模式,所述AI模型参数的指示方式。
  2. 根据权利要求1所述的方法,其中,所述更新模式包括以下之一:第一模式、第二模式、第三模式、第四模式;
    其中,所述第一模式包括:更新第一参数子集中的所有参数,所述第一参数子集是所述AI模型参数的子集,所述第一参数子集是预配置或者协议约定的或者网络侧指示的;
    所述第二模式包括:更新第二参数子集中的所有参数,所述第二参数子集是所述第一参数子集的子集;
    所述第三模式包括:更新所述AI模型的所有参数;
    所述第四模式包括:更新第三参数子集,所述第三参数子集是所述AI模型参数的子集。
  3. 根据权利要求1所述的方法,其中,所述指示方式包括以下之一:
    第一信息,所述第一信息用于指示所述AI模型参数在AI模型中的位置;
    第二信息,所述第二信息用于指示所述AI模型参数的发送顺序。
  4. 根据权利要求3所述的方法,其中,所述AI模型为神经网络,所述神经网络包括输入层、一个或多个隐藏层和输出层,所述输入层、一个或多个隐藏层、输出层,以及各层的神经元组成多行多列的结构,所述神经网络中神经元的参数的发送顺序是根据所述神经元在所述神经网络中的位置确定的。
  5. 根据权利要求4所述的方法,其中,所述神经网络中神经元的参数的发送顺序,包括:
    第一列所有神经元的参数,第二列所有神经元的参数,以此类推,直至最后一列所有神经元的参数;或,
    第一行所有神经元的参数,第二行所有神经元的参数,以此类推,直至最后一行所有神经元的参数。
  6. 根据权利要求5所述的方法,其中,
    所述第一列对应所述输入层,所述最后一列对应所述输出层,或者,所述第一列对应所述输出层,所述最后一列对应所述输入层;
    或者,
    所述第一行对应所述输入层,所述最后一行对应所述输出层;或者,所述第一行对应所述输出层,所述最后一行对应所述输入层。
  7. 根据权利要求5所述的方法,其中,在所述神经网络中神经元的参数的发送顺序包括第一列所有神经元的参数,第二列所有神经元的参数,以此类推,直至最后一列所有神经元的参数的情况下,所述神经网络中神经元的参数的发送顺序还包括以下之一:
    每一列中神经元的参数按照从上往下的神经元的顺序发送;
    每一列中神经元的参数按照从下往上的神经元的顺序发送;
    每一列中神经元的参数的发送顺序相同;
    相邻列的神经元的参数的发送顺序相反。
  8. 根据权利要求5所述的方法,其中,所述第一行是所述神经网络中最上方一行,所述最后一行是所述神经网络中最下方一行;或者,所述第一行是所述神经网络中最下方一行,所述最后一行是所述神经网络中最上方一行。
  9. 根据权利要求5所述的方法,其中,在所述神经网络中神经元的参数的发送顺序包括第一行所有神经元的参数,第二行所有神经元的参数,以此类推,直至最后一行所有神经元的参数的情况下,所述神经网络中神经元的参数的发送顺序还包括以下之一:
    每一行中神经元的参数按照从左往右的神经元的顺序发送;
    每一行中神经元的参数按照从右往左的神经元的顺序发送;
    每一行中神经元的参数的发送顺序相同;
    相邻行的神经元的参数的发送顺序相反。
  10. 根据权利要求9所述的方法,其中,在每层的神经元数目不是完全相同的情况下,每一行都按最多神经元数量的行发送所述神经元的参数,缺少神经元的层补预设值,或者,每一行都按实际神经元数量发送所述神经元的参数。
  11. 根据权利要求3所述的方法,其中,所述AI模型参数的发送顺序是预配置的或协议约定的或者网络侧指示的。
  12. 根据权利要求3所述的方法,其中,所述AI模型参数包括乘性系数和加性系数,所述乘性系数的发送优先级高于所述加性系数的发送优先级;
    或者,
    所述加性系数的发送优先级高于所述乘性系数的发送优先级。
  13. 根据权利要求1所述的方法,其中,所述AI模型的量化等级和/或压缩方式是预配置的或者协议约定的或者网络侧指示的。
  14. 根据权利要求1所述的方法,其中,更新后的AI模型的模型ID与更新前的AI模型的模型ID相同,不保存所述更新前的AI模型;
    或者,
    更新后的AI模型的模型ID与更新前的AI模型的模型ID不同,保存所述更新前的AI模型,且所述更新前的AI模型的模型ID不变。
  15. 根据权利要求1所述的方法,其中,所述AI模型包括第一功能模块,所述第一功能模块用于以下一项或多项:
    信号处理;
    信道传输、信道接收、信道解调或信道发送;
    获取信道状态信息;
    波束管理;
    信道预测;
    干扰抑制;
    定位;
    预测或管理高层业务和/或高层参数的预测;
    解析控制信令。
  16. 根据权利要求1所述的方法,其中,所述第一节点为第一网络侧设备或第一终端,所述第二节点为第二网络侧设备或第二终端。
  17. 一种更新AI模型参数的方法,包括:
    第二节点接收第一节点发送的与AI模型参数相关的第一信息;
    所述第二节点根据所述第一信息更新所述AI模型参数;
    其中,所述第一信息包括以下至少一项:所述AI模型参数的更新模式,所述AI模型参数的指示方式。
  18. 根据权利要求17所述的方法,其中,所述更新模式包括以下之一:第一模式、第二模式、第三模式、第四模式;
    其中,所述第一模式包括:更新第一参数子集中的所有参数,所述第一参数子集是所述AI模型参数的子集,所述第一参数子集是预配置或者协议约定的或者网络侧指示的;
    所述第二模式包括:更新第二参数子集中的所有参数,所述第二参数子集是所述第一参数子集的子集,所述第一参数子集是所述AI模型参数的子集,所述第一参数子集是预先指示或协议约定的;
    所述第三模式包括:更新所述AI模型的所有参数;
    所述第四模式包括:更新第三参数子集,所述第三参数子集是所述AI模型参数的子集。
  19. 根据权利要求17所述的方法,其中,所述指示方式包括以下之一:
    第一信息,所述第一信息用于指示所述AI模型参数在所述AI模型中的位置;
    第二信息,所述第二信息用于指示所述AI模型参数的发送顺序。
  20. 根据权利要求19所述的方法,其中,所述AI模型为神经网络,所述神经网络包括输入层、一个或多个隐藏层和输出层,所述输入层、一个或多个隐藏层、输出层,以及各层的神经元组成多行多列的结构,所述神经网络中神经元的参数的发送顺序是根据所述神经元在所述神经网络中的位置确定的。
  21. 根据权利要求20所述的方法,其中,所述神经网络中神经元的参数的发送顺序,包括:
    第一列所有神经元的参数,第二列所有神经元的参数,以此类推,直至最后一列所有神经元的参数;
    第一行所有神经元的参数,第二行所有神经元的参数,以此类推,直至最后一行所有神经元的参数。
  22. 根据权利要求21所述的方法,其中,
    所述第一列对应所述输入层,所述最后一列对应所述输出层,或者,所述第一列对应所述输出层,所述最后一列对应所述输入层;
    或者,
    所述第一行对应所述输入层,所述最后一行对应所述输出层;或者,所述第一行对应所述输出层,所述最后一行对应所述输入层。
  23. 根据权利要求21所述的方法,其中,在所述神经网络中神经元的参数的发送顺序包括第一列所有神经元的参数,第二列所有神经元的参数,以此类推,直至最后一列所有神经元的参数的情况下,所述神经网络中神经元的参数的发送顺序还包括以下之一:
    每一列中神经元的参数按照从上往下的神经元的顺序发送;
    每一列中神经元的参数按照从下往上的神经元的顺序发送;
    每一列中神经元的参数的发送顺序相同;
    相邻列的神经元的参数的发送顺序相反。
  24. 根据权利要求21所述的方法,其中,所述第一行是所述神经网络中最上方一行,所述最后一行是所述神经网络中最下方一行;或者,所述第一行是所述神经网络中最下方一行,所述最后一行是所述神经网络中最上方一行。
  25. 根据权利要求21所述的方法,其中,在所述神经网络中神经元的参数的发送顺序包括第一行所有神经元的参数,第二行所有神经元的参数,以此类推,直至最后一行所有神经元的参数的情况下,所述神经网络中神经元的参数的发送顺序还包括以下之一:
    每一行中神经元的参数按照从左往右的神经元的顺序发送;
    每一行中神经元的参数按照从右往左的神经元的顺序发送;
    每一行中神经元的参数的发送顺序相同;
    相邻行的神经元的参数的发送顺序相反。
  26. 根据权利要求25所述的方法,其中,在每层的神经元数目不是完全相同的情况下,每一行都按最多神经元数量的行发送所述神经元的参数,缺少神经元的层补预设值,或者,每一行都按实际神经元数量发送所述神经元的参数。
  27. 根据权利要求19所述的方法,其中,所述AI模型参数的发送顺序是预配置的或协议约定的或者网络侧指示的。
  28. 根据权利要求19所述的方法,其中,所述AI模型中乘性系数的发送优先级高于加性系数的发送优先级;
    或者,
    所述AI模型中加性系数的发送优先级高于乘性系数的发送优先级。
  29. 根据权利要求17所述的方法,其中,所述AI模型的量化等级和/或压缩方式是预配置的或者协议约定的或者网络侧指示的。
  30. 根据权利要求17所述的方法,其中,更新后的AI模型的模型ID与更新前的AI模型的模型ID相同,不保存所述更新前的AI模型;
    或者,
    更新后的AI模型的模型ID与更新前的AI模型的模型ID不同,保存所述更新前的AI模型,且所述更新前的AI模型的模型ID不变。
  31. 根据权利要求17所述的方法,其中,所述AI模型包括第一功能模块,所述第一功能模块用于以下一项或多项:
    信号处理;
    信道传输、信道接收、信道解调或信道发送;
    获取信道状态信息;
    波束管理;
    信道预测;
    干扰抑制;
    定位;
    预测或管理高层业务和/或高层参数的预测;
    解析控制信令。
  32. 根据权利要求17所述的方法,其中,所述第二节点为第二网络侧设备或第二终端,所述第一节点为第一网络侧设备或第一终端。
  33. 一种更新AI模型参数的装置,包括:
    第一发送模块,用于向第二节点发送与AI模型参数相关的第一信息;
    其中,所述第一信息包括以下至少一项:所述AI模型参数的更新模式,所述AI模型参数的指示方式。
  34. 一种更新AI模型参数的装置,包括:
    第一接收模块,用于接收第一节点发送的与AI模型参数相关的第一信息;
    更新模块,用于根据所述第一信息更新所述AI模型参数;
    其中,所述第一信息包括以下至少一项:所述AI模型参数的更新模式,所述AI模型参数的指示方式。
  35. 一种通信设备,包括处理器,存储器及存储在所述存储器上并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32中任一项所述的方法的步骤。
  36. 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32中任一项所述的方法的步骤。
PCT/CN2023/119938 2022-09-26 2023-09-20 更新ai模型参数的方法、装置及通信设备 Ceased WO2024067280A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211177497.5A CN117834427A (zh) 2022-09-26 2022-09-26 更新ai模型参数的方法、装置及通信设备
CN202211177497.5 2022-09-26

Publications (1)

Publication Number Publication Date
WO2024067280A1 true WO2024067280A1 (zh) 2024-04-04

Family

ID=90476175

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/119938 Ceased WO2024067280A1 (zh) 2022-09-26 2023-09-20 更新ai模型参数的方法、装置及通信设备

Country Status (2)

Country Link
CN (1) CN117834427A (zh)
WO (1) WO2024067280A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN121396804A (zh) * 2024-07-22 2026-01-23 维沃移动通信有限公司 模型参数传输方法、装置及通信设备
CN121925875A (zh) * 2024-08-09 2026-04-24 北京小米移动软件有限公司 通信方法、终端、网络设备、通信装置及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108924910A (zh) * 2018-07-25 2018-11-30 Oppo广东移动通信有限公司 Ai模型的更新方法及相关产品
US20200151558A1 (en) * 2018-11-13 2020-05-14 Gyrfalcon Technology Inc. Systems and methods for updating an artificial intelligence model by a subset of parameters in a communication system
CN114091679A (zh) * 2020-08-24 2022-02-25 华为技术有限公司 一种更新机器学习模型的方法及通信装置
CN114363921A (zh) * 2020-10-13 2022-04-15 维沃移动通信有限公司 Ai网络参数的配置方法和设备
CN114519435A (zh) * 2022-02-14 2022-05-20 维沃移动通信有限公司 模型参数更新方法、模型参数更新装置和电子设备

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108924910A (zh) * 2018-07-25 2018-11-30 Oppo广东移动通信有限公司 Ai模型的更新方法及相关产品
US20200151558A1 (en) * 2018-11-13 2020-05-14 Gyrfalcon Technology Inc. Systems and methods for updating an artificial intelligence model by a subset of parameters in a communication system
CN114091679A (zh) * 2020-08-24 2022-02-25 华为技术有限公司 一种更新机器学习模型的方法及通信装置
CN114363921A (zh) * 2020-10-13 2022-04-15 维沃移动通信有限公司 Ai网络参数的配置方法和设备
CN114519435A (zh) * 2022-02-14 2022-05-20 维沃移动通信有限公司 模型参数更新方法、模型参数更新装置和电子设备

Also Published As

Publication number Publication date
CN117834427A (zh) 2024-04-05

Similar Documents

Publication Publication Date Title
US20240224082A1 (en) Parameter selection method, parameter configuration method, terminal, and network side device
US20240323741A1 (en) Measurement method and apparatus, device, and storage medium
WO2024067281A1 (zh) Ai模型的处理方法、装置及通信设备
WO2023186099A1 (zh) 信息反馈方法、装置及设备
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
US20250021841A1 (en) Sample determining method and apparatus, and device
US20250310756A1 (en) Ai computing power reporting method, terminal, and network-side device
WO2023179540A1 (zh) 信道预测方法、装置及无线通信设备
WO2024099091A1 (zh) 波束预测方法、装置、终端、网络侧设备及存储介质
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
WO2024093713A1 (zh) 资源配置方法、装置、通信设备及可读存储介质
WO2025146039A1 (zh) 信息处理方法、装置及通信设备
WO2024032695A1 (zh) Csi预测处理方法、装置、通信设备及可读存储介质
CN121510038A (zh) Ai模型传输的指示方法、装置、通信设备及存储介质
WO2025140604A1 (zh) 信息上报方法、装置、第一设备及第二设备
WO2026032266A1 (zh) 无线通信方法、装置及设备
WO2025201273A1 (zh) 模型处理方法、装置、通信设备及存储介质
WO2024067665A1 (zh) Csi预测处理方法、装置、通信设备及可读存储介质
CN121510039A (zh) 关联标识的关联范围的确定方法、装置、通信设备及介质
CN121485735A (zh) Csi上报、接收方法、装置、设备、可读存储介质及计算机程序产品
CN120238893A (zh) 信息指示方法、装置、通信设备及可读存储介质
CN120238974A (zh) 功能切换方法、装置、设备及可读存储介质
WO2025140233A1 (zh) Csi预测方法、csi预测结果的监控方法、装置、设备及可读存储介质
CN122001958A (zh) 传输方法、装置、设备、可读存储介质及计算机程序产品

Legal Events

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

Ref document number: 23870512

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 23870512

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 04.09.2025)