[go: up one dir, main page]

WO2025140663A1 - Procédé, appareil et système d'acquisition de données d'un modèle - Google Patents

Procédé, appareil et système d'acquisition de données d'un modèle Download PDF

Info

Publication number
WO2025140663A1
WO2025140663A1 PCT/CN2024/143477 CN2024143477W WO2025140663A1 WO 2025140663 A1 WO2025140663 A1 WO 2025140663A1 CN 2024143477 W CN2024143477 W CN 2024143477W WO 2025140663 A1 WO2025140663 A1 WO 2025140663A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
node
data
original data
model
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.)
Pending
Application number
PCT/CN2024/143477
Other languages
English (en)
Chinese (zh)
Inventor
陈家璇
孙琰
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.)
Huawei Technologies Co Ltd
Original Assignee
Huawei Technologies 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 Huawei Technologies Co Ltd filed Critical Huawei Technologies Co Ltd
Publication of WO2025140663A1 publication Critical patent/WO2025140663A1/fr
Pending legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/098—Distributed learning, e.g. federated learning

Definitions

  • the present application relates to the field of artificial intelligence (AI), and in particular to a model data acquisition method, device and system.
  • AI artificial intelligence
  • model training or fine-tuning requires sufficient training data for the corresponding scenarios to ensure the convergence performance of the model, but the measured data obtained on the device side is limited, and it is difficult to ensure model performance only by training with data collected by the device; on the other hand, model training or fine-tuning requires data features that are rich enough to ensure the generalization and robustness of the model, but the measured data obtained on the device side is limited, resulting in limited diversity of data features.
  • the central node receives and merges the original local data of multiple devices as a training set to increase the amount of training data and data diversity of the training set. Then, the central node uses the merged training set to perform model training or model update.
  • the trained model is deployed on the device side for model inference. Among them, the central node only uses the original local data of multiple devices as the input of the generation model to obtain synthetic data, and the quality of the synthetic data is not high.
  • the present application discloses a model data acquisition method, device and system, which can improve the quality of synthetic data.
  • the first node or a circuit used for the first node such as a chip, generates a plurality of synthetic data based on the acquired original data and characteristic information of the original data.
  • the first node further sends second information to the second node.
  • the second information indicates a processing method for the second node to obtain the first information.
  • the processing method for the second node to obtain the first information includes a method for acquiring feature information of the original data.
  • the first node determines the number of similar synthetic data to be generated according to the priority or probability of each original data, so that the distribution of the synthetic data is consistent with expectations.
  • the processing method for the second node to obtain the first information also includes a method for obtaining the data label of the original data.
  • the second node Based on the method for obtaining the data label of the original data, the second node processes the original data to obtain the data label of the original data.
  • the first node further sends third information to the second node, where the third information indicates a reporting configuration of the first information.
  • the first node sends the third information to the second node so that the second node reports based on the reporting configuration of the first information indicated in the third information, so that the data provider and the synthesized data end separated from each other can align their understanding of the data and improve the quality of the synthesized data.
  • the reporting configuration includes at least one of the following: the type of the first information, and the data volume of the first information.
  • the first node sends a reporting configuration of the type of first information and/or a reporting configuration of the data amount of the first information to the second node, so that the second node sends the corresponding type of first information or the corresponding data amount of first information to the first node, which helps to improve the quality of the synthesized data.
  • the first node further sends fourth information to the second node.
  • the fourth information indicates that the first node has the ability to jointly process feature information of multiple original data of the same group.
  • the joint processing may be, for example, weighted summing of feature information of multiple original data of the same group.
  • the first node sends the fourth information to the second node so that the second node can divide the original data that can be jointly processed into the same group, while dividing the original data that cannot be jointly processed into the same group, thereby avoiding the fusion of original data from different groups to generate unexpected synthetic data.
  • the first information further includes the following type of information: group information of the original data.
  • the second node reports the group information of the original data to the first node, and then the first node can jointly process the feature information of multiple original data in the same group.
  • the first node further sends at least one subset of the plurality of synthesized data to the second node.
  • the first node further receives fifth information from the second node.
  • the fifth information indicates an evaluation result of at least one synthesized data in the at least one subset.
  • the evaluation result indicates at least one synthesized data that needs to be eliminated in the at least one subset, or indicates at least one synthesized data that does not need to be eliminated in the at least one subset.
  • the evaluation result is used to filter the plurality of synthetic data to obtain at least one filtered synthetic data, wherein the at least one filtered synthetic data is used for processing the model.
  • the at least one synthesized data after screening is determined based on the distance between the first synthesized data to be eliminated and other synthesized data in the multiple synthesized data, and the first synthesized data to be eliminated is the synthesized data that needs to be eliminated determined based on the fifth information.
  • an embodiment of the present application provides a model data acquisition method, which is executed by a first node or a circuit for the first node.
  • the method includes: the first node receives sixth information from the second node.
  • the sixth information includes the following types of information: original data.
  • the first node generates multiple synthetic data based on the sixth information and the characteristic information of the original data.
  • the multiple synthetic data are used for model processing. The following description is based on the first node as the execution subject. It can be understood that the first node can also be replaced by a circuit for the first node, such as a chip.
  • the first node further receives eleventh information from the third node, where the eleventh information includes the following type of information: characteristic information of the original data.
  • the first node further receives ninth information from the second node, the ninth information indicating a method for the first node to obtain the characteristic information of the original data. Further, the first node processes the original data based on the ninth information to obtain the characteristic information of the original data.
  • the first node further receives ninth information from the third node, the ninth information indicating a method for the first node to obtain the characteristic information of the original data. Further, the first node processes the original data based on the ninth information to obtain the characteristic information of the original data.
  • the characteristic information of the original data is determined based on a method for acquiring the characteristic information of the original data, and the method for acquiring the characteristic information of the original data is determined by the first node.
  • the sixth information further includes at least one of the following types of information: the priority of the original data, or a data label of the original data.
  • the first node determines the number of similar synthetic data to be generated according to the priority or probability of each original data, so that the distribution of the synthetic data is consistent with expectations.
  • the first node also sends eighth information to the second node, where the eighth information indicates a processing method for the second node to obtain the sixth information, and the processing method for the second node to obtain the sixth information includes a method for obtaining a data label of the original data.
  • the second node Based on the method for obtaining the data label of the original data, the second node processes the original data to obtain the data label of the original data.
  • the first node further sends seventh information to the second node, where the seventh information indicates a reporting configuration of the sixth information.
  • the first node sends the seventh information to the second node so that the second node reports based on the reporting configuration of the sixth information indicated in the seventh information.
  • the data provider and the synthesized data end separated from each other can align their understanding of the data, thereby improving the quality of the synthesized data.
  • the reporting configuration includes at least one of the following: the type of the sixth information, or the data volume of the sixth information.
  • the first node sends a reporting configuration of the type of sixth information and/or a reporting configuration of the data amount of the sixth information to the second node, so that the second node sends the corresponding type of sixth information, or the corresponding data amount of sixth information to the first node, which helps to improve the quality of the synthesized data.
  • the first node further sends fourth information to the second node.
  • the fourth information indicates that the first node has the ability to jointly process feature information of multiple original data of the same group.
  • the joint processing may be, for example, weighted summing of feature information of multiple original data of the same group.
  • the first node sends the fourth information to the second node so that the second node can divide the original data that can be jointly processed into the same group, while dividing the original data that cannot be jointly processed into the same group, thereby avoiding the fusion of original data from different groups to generate unexpected synthetic data.
  • the sixth information further includes the following type of information: group information of the original data.
  • the second node reports the group information of the original data to the first node, and then the first node can jointly process the feature information of multiple original data in the same group.
  • the first node further sends at least one subset of the plurality of synthesized data to the second node.
  • the first node further receives fifth information from the second node.
  • the fifth information indicates an evaluation result of at least one synthesized data in the at least one subset.
  • the evaluation result indicates at least one synthesized data that needs to be eliminated in the at least one subset, or indicates at least one synthesized data that does not need to be eliminated in the at least one subset.
  • the evaluation result is used to filter the plurality of synthetic data to obtain at least one filtered synthetic data, wherein the at least one filtered synthetic data is used for processing the model.
  • the at least one synthesized data after screening is determined based on the distance between the first synthesized data to be eliminated and other synthesized data in the multiple synthesized data, and the first synthesized data to be eliminated is the synthesized data that needs to be eliminated determined based on the fifth information.
  • an embodiment of the present application provides a model data acquisition method, which is executed by a second node or a circuit for a second node.
  • the method includes: sending first information to a first node.
  • the first information includes the following types of information: original data and characteristic information of the original data.
  • the following description is based on the second node as the execution subject. It can be understood that the second node can also be replaced by a circuit for the second node, such as a chip.
  • the second node sends original data and characteristic information of the original data to the first node, and the original data and the characteristic information of the original data are used to generate multiple synthetic data.
  • synthetic data that meets diversity requirements and has the same data distribution as the measured data can be generated, which can improve the quality of the synthetic data and thus help improve the performance of model training or updating.
  • the second node further receives second information from the first node.
  • the second information indicates a processing method by which the second node obtains the first information.
  • the processing method by which the second node obtains the first information includes a method for obtaining characteristic information of the original data. Further, the second node obtains the first information based on the second information.
  • the first information further includes at least one of the following types of information: a data label of the original data, or a priority of the original data.
  • the processing method also includes a method for acquiring data labels of the original data.
  • the second node further receives third information from the first node, where the third information indicates a reporting configuration of the first information.
  • the reporting configuration includes at least one of the following: the type of the first information, and the data volume of the first information.
  • the second node further receives fourth information from the first node, where the fourth information indicates that the first node has the ability to jointly process feature information of multiple original data of the same group.
  • the first information further includes the following type of information: group information of the original data.
  • the second node further receives at least one subset of the multiple synthetic data from the first node.
  • the second node further sends fifth information to the first node, where the fifth information indicates an evaluation result of at least one synthetic data in the at least one subset.
  • the evaluation result indicates at least one synthetic data that needs to be eliminated from the at least one subset, or indicates at least one synthetic data that does not need to be eliminated from the at least one subset.
  • the evaluation result is determined based on a distance between at least one synthetic data in the at least one subset and the local data.
  • the evaluation result is used to filter the multiple synthetic data to obtain at least one filtered synthetic data; wherein the at least one filtered synthetic data is used for model processing.
  • an embodiment of the present application provides a model data acquisition method, which is executed by a second node or a circuit for a second node.
  • the method includes: the second node sends sixth information to the first node.
  • the sixth information includes the following types of information: original data.
  • the second node also receives eighth information from the first node, where the eighth information indicates a processing method for the second node to obtain the sixth information, and the processing method for the second node to obtain the sixth information includes a method for obtaining a data label of the original data.
  • the second node further receives at least one subset of the multiple synthetic data from the first node.
  • the second node further sends fifth information to the first node, where the fifth information indicates an evaluation result of at least one synthetic data in the at least one subset.
  • the evaluation result indicates at least one synthetic data that needs to be eliminated from the at least one subset, or indicates at least one synthetic data that does not need to be eliminated from the at least one subset.
  • the evaluation result is used to filter the multiple synthetic data to obtain at least one filtered synthetic data; wherein the at least one filtered synthetic data is used for model processing.
  • an embodiment of the present application provides a model data acquisition method, which is performed by a third node or a circuit for a third node.
  • the method includes: sending eleventh information to a first node, the eleventh information including the following types of information: characteristic information of the original data.
  • the first node obtains synthetic data based on the characteristic information of the original data, that is, the characteristic information of the original data is used to generate the synthetic data.
  • an embodiment of the present application provides a model data acquisition method, which is executed by a third node or a circuit for a third node.
  • the method includes: sending ninth information to a first node, wherein the ninth information indicates a method for the first node to obtain characteristic information of the original data.
  • the first node processes the original data based on the ninth information to obtain characteristic information of the original data, that is, the method for obtaining characteristic information of the original data is used to obtain characteristic information of the original data.
  • the present application provides a model data acquisition device, which may include a transceiver module and a processing module, as follows:
  • a transceiver module configured to receive first information from a second node, wherein the first information includes the following types of information: original data and characteristic information of the original data;
  • a processing module is used to generate a plurality of synthetic data based on the first information, and the plurality of synthetic data are used for processing the model.
  • the transceiver module is further used to send second information to the second node, where the second information indicates a processing method for the second node to obtain the first information, and the processing method for the second node to obtain the first information includes a method for obtaining characteristic information of the original data.
  • the first information further includes the following types of information: a data label of the original data and/or a priority of the original data.
  • the processing method for the second node to obtain the first information also includes a method for obtaining the data label of the original data.
  • the transceiver module is further configured to send third information to the second node, where the third information indicates a reporting configuration of the first information.
  • the reporting configuration includes at least one of the following: a type of the first information, and a data volume of the first information.
  • the transceiver module is further configured to send fourth information to the second node, where the fourth information indicates that the device has the ability to jointly process feature information of multiple original data of the same group.
  • the first information further includes the following type of information: group information of the original data.
  • the transceiver module is further configured to send at least one subset of the plurality of synthesized data to the second node;
  • the at least one synthesized data after screening is determined based on the distance between the first synthesized data to be eliminated and other synthesized data in the multiple synthesized data, and the first synthesized data to be eliminated is the synthesized data that needs to be eliminated determined based on the fifth information.
  • the transceiver module is used to receive sixth information from the second node, where the sixth information includes the following types of information: original data.
  • the first information further includes the following types of information: a data label of the original data and/or a priority of the original data.
  • the transceiver module is further configured to receive at least one subset of the plurality of synthesized data from the first node;
  • the transceiver module is used to send sixth information to the first node.
  • the sixth information includes the following types of information: original data.
  • the transceiver module is used to send eleventh information to the first node, where the eleventh information includes the following types of information: characteristic information of original data.
  • the present application provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement a method provided in any possible implementation of the first aspect, or a method provided in any possible implementation of the second aspect, or a method provided in any possible implementation of the third aspect.
  • FIG6 is a schematic diagram of a flow chart of a method for acquiring model data provided in an embodiment of the present application.
  • FIG11 is a schematic diagram of the structure of a model data acquisition device provided in an embodiment of the present application.
  • FIG. 12 is a schematic diagram of the structure of another model data acquisition device provided in an embodiment of the present application.
  • the present disclosure relates to at least one (item) as follows, indicating one (item) or more (items). More than one (item) refers to two (items) or more than two (items).
  • "And/or" describes the association relationship of associated objects, indicating that three relationships may exist.
  • a and/or B can represent: A exists alone, A and B exist at the same time, and B exists alone.
  • the character "/” generally indicates that the objects associated before and after are in an "or” relationship.
  • first, second, etc. may be used to describe each object in the present disclosure, these objects should not be limited to these terms. These terms are only used to distinguish each object from each other.
  • the communication system can be a fifth generation (5G) or new radio (NR) system, a long term evolution (LTE) system, an LTE frequency division duplex (FDD) system, an LTE time division duplex (TDD) system, a wireless local area network (WLAN) system, a satellite communication system, a future communication system such as a sixth generation (6G) mobile communication system, or a fusion system of multiple systems.
  • 5G fifth generation
  • NR new radio
  • LTE long term evolution
  • FDD LTE frequency division duplex
  • TDD LTE time division duplex
  • WLAN wireless local area network
  • 6G sixth generation
  • the technical solution provided by the present application can also be applied to device to device (D2D) communication, vehicle to everything (V2X) communication, machine to machine (M2M) communication, machine type communication (MTC), and Internet of things (IoT) communication system or other communication systems.
  • V2X vehicle to everything
  • M2M machine to machine
  • MTC machine type communication
  • IoT Internet of things
  • a device in a communication system can send a signal to another device or receive a signal from another device.
  • the signal may include information, signaling, or data, etc.
  • the device may also be replaced by an entity, a network entity, a network element, a communication device, a communication module, a node, a communication node, etc.
  • the present disclosure is described by taking the device as an example.
  • the communication system may include at least one terminal device and at least one access network device.
  • the access network device may send a downlink signal to the terminal device, and/or the terminal device may send an uplink signal to the access network device.
  • multiple terminal devices may also send signals to each other, that is, the signal sending device and the signal receiving device may both be terminal devices.
  • the wireless communication system may include multiple network devices (also referred to as access network devices) at the same time, or may include multiple communication devices at the same time.
  • a network device may serve one or more communication devices at the same time.
  • a communication device may also access one or more network devices at the same time.
  • the embodiment of the present application does not limit the number of communication devices and network devices included in the wireless communication system.
  • the network device can be an entity on the network side for transmitting or receiving signals.
  • the network device can be an access device for the communication device to access the wireless communication system by wireless means, such as the network device can be a base station.
  • the base station can broadly cover the following various names, or be replaced with the following names, such as: Node B (NodeB), evolved NodeB (eNB), next generation NodeB (gNB), access network equipment in open radio access network (O-RAN), relay station, access point, transmission point (transmitting and receiving point, TRP), transmitting point (transmitting point, TP), master station (Master eNodeB, MeNB), secondary eNodeB (SeNB), multi-standard radio (Multi-standard radio o, MSR) node, home base station, network controller, access node, wireless node, access point (AP), transmission node, transceiver node, baseband unit (BBU), remote radio unit (RRU), active antenna unit (AAU), remote radio head (RRH), central unit (CU), distributed
  • the base station can be a macro base station, a micro base station, a relay node, a donor node or the like, or a combination thereof.
  • the network device can also refer to a communication module, a modem or a chip used to be set in the aforementioned device or apparatus.
  • the network device may also be a mobile switching center and a device that performs base station functions in device-to-device (D2D), vehicle-to-everything (V2X), and machine-to-machine (M2M) communications, a network-side device in a 6G network, and a device that performs base station functions in future communication systems.
  • the network device may support networks with the same or different access technologies. The embodiments of the present application do not limit the specific technology and specific device form used by the network device.
  • the network equipment may be fixed or mobile.
  • base stations 110a, 110b are stationary and are responsible for wireless transmission and reception in one or more cells from the communication device 120.
  • the helicopter or drone 120i shown in FIG. 1 may be configured to act as a mobile base station, and one or more cells may move according to the location of the mobile base station 120i.
  • the helicopter or drone (120i) may be configured to act as a communication device that communicates with the base station 110b.
  • the communication device used to implement the above access network function can be an access network device, or a network device with some functions of accessing the network, or a device capable of supporting the implementation of the access network function, such as a chip system, a hardware circuit, a software module, or a hardware circuit plus a software module, which can be installed in the access network device or used in combination with the access network device.
  • the communication device used to implement the access network device function is an access network device for description as an example.
  • the communication device can be an entity on the user side for receiving or transmitting signals, such as a mobile phone.
  • the communication device can be used to connect people, objects and machines.
  • the communication device can communicate with one or more core networks through a network device.
  • the communication device includes a handheld device with a wireless connection function, other processing devices connected to a wireless modem, or a vehicle-mounted device.
  • the communication device can be a portable, pocket-sized, handheld, computer-built-in or vehicle-mounted mobile device.
  • the communication device 120 can be widely used in various scenarios, such as cellular communication, device-to-device D2D, vehicle-to-all V2X, peer-to-peer (P2P), machine-to-machine M2M, machine type communication MTC, Internet of Things IoT, virtual reality (VR), augmented reality (AR), industrial control, autonomous driving, telemedicine, smart grid, smart furniture, smart office, smart wear, smart transportation, smart city, drone, robot, remote sensing, passive sensing, positioning, navigation and tracking, autonomous delivery and mobility, etc.
  • cellular communication device-to-device D2D, vehicle-to-all V2X, peer-to-peer (P2P), machine-to-machine M2M, machine type communication MTC, Internet of Things IoT, virtual reality (VR), augmented reality (AR), industrial control, autonomous driving, telemedicine, smart grid, smart furniture, smart office, smart wear, smart transportation, smart city, drone, robot, remote sensing, passive sensing, positioning, navigation and tracking, autonomous delivery and mobility
  • Some examples of communication devices 120 are: 3GPP standard user equipment (UE), fixed equipment, mobile devices, handheld devices, wearable devices, cellular phones, smart phones, Session Initialization Protocol (SIP) phones, laptops, personal computers, smart books, vehicles, satellites, Global Positioning System (GPS) devices, target tracking devices, drones, helicopters, aircraft, ships, remote control devices, smart home devices, industrial equipment, personal communication service (PCS) phones, wireless local loop (WLL) stations, personal digital assistants (PDAs), etc.
  • the communication device 120 may be a wireless device in the above-mentioned scenarios or a device used to be set in a wireless device, for example, a communication module, a modem or a chip in the above-mentioned device.
  • the communication device may also be referred to as a terminal, a terminal device, a user equipment UE, a mobile station (MS), a mobile terminal (MT), or the like.
  • the communication device may also be referred to as a terminal, a terminal device, a user equipment UE, a mobile station (MS), a mobile terminal (MT), or the like.
  • the communication device may also be referred to as a communication device in a future wireless communication system.
  • the communication device can be used in a dedicated network device or a general device. The embodiments of the present application do not limit the specific technology and specific device form used by the communication device.
  • the communication device used to implement the functions of the communication device may be a terminal device, or a terminal device having some of the functions of the above communication device, or a device capable of supporting the implementation of the functions of the above communication device, such as a chip system, which may be installed in the terminal device or used in combination with the terminal device.
  • the chip system may be composed of a chip, or may include a chip and other discrete devices.
  • the communication device is described as a terminal device or UE as an example.
  • a wireless communication system is usually composed of cells, and a base station provides management of the cell.
  • the base station provides communication services to multiple mobile stations (MS) in the cell.
  • the base station includes a baseband unit (BBU) and a remote radio unit (RRU).
  • BBU baseband unit
  • RRU remote radio unit
  • the BBU and RRU can be placed in different places, for example: the RRU is remote and placed in an area with high traffic volume, and the BBU is placed in a central computer room.
  • the BBU and RRU can also be placed in the same computer room.
  • the BBU and RRU can also be different components under one rack.
  • a cell can correspond to one carrier or component carrier.
  • the present disclosure can be applied between a network device and a communication device, between a network device and a network device, or between a communication device and a communication device, that is, between a primary device and a secondary device.
  • the primary device can be a network device or a communication device.
  • the secondary device can be another network device or a communication device.
  • the secondary device can be another communication device.
  • AI nodes may also be introduced into the network.
  • the method provided by the present invention can be used for communication between access network equipment and terminal equipment, and can also be used for communication between other communication equipment, such as communication between macro base stations and micro base stations in a wireless backhaul link, and communication between two terminal devices in a side link (SL), without limitation.
  • other communication equipment such as communication between macro base stations and micro base stations in a wireless backhaul link, and communication between two terminal devices in a side link (SL), without limitation.
  • SL side link
  • An AI model is an algorithm or computer program that can implement AI functions.
  • the AI model represents the mapping relationship between the input and output of the model.
  • the type of AI model can be a neural network, linear regression model, decision tree model, support vector machine (SVM), Bayesian network, Q learning model or other machine learning (ML) model.
  • An autoencoder is a neural network for unsupervised learning. Its characteristic is that it uses input data as labels, so an autoencoder can also be understood as a neural network for self-supervised learning.
  • An autoencoder can be used for data compression and recovery. For example, the encoder in an autoencoder can compress (encode) data A to obtain data B; the decoder in an autoencoder can decompress (decode) data B to recover data A. Alternatively, it can be understood that the decoder is the inverse operation of the encoder.
  • the AI model in the embodiment of the present application may be a single-end model, which may be deployed on a terminal device or a network device.
  • ground truth usually refers to data that is believed to be accurate or real.
  • the training data set is used for training the AI model.
  • the training data set may include the input of the AI model, or the input and target output of the AI model.
  • the training data set includes one or more training data.
  • the training data may include training samples input to the AI model, or may include the target output of the AI model.
  • the target output may also be referred to as a label, sample label, or label sample.
  • the label is the true value.
  • training data sets can include simulation data collected through simulation platforms, experimental data collected in experimental scenarios, or measured data collected in actual communication networks. Due to differences in the geographical environment and channel conditions in which the data is generated, such as indoor and outdoor conditions, mobile speeds, frequency bands, or antenna configurations, the collected data can be classified when acquiring the data. For example, data with the same channel propagation environment and antenna configuration can be grouped together.
  • Model training is essentially learning some of its features from the training data.
  • AI models such as neural network models
  • the AI model is a neural network, and adjusting the model parameters of the neural network includes adjusting at least one of the following parameters: the number of layers, width, weights of neurons, or parameters in the activation function of the neurons of the neural network.
  • Inference data can be used as input to the trained AI model for inference of the AI model.
  • the inference data is input into the AI model, and the corresponding output is the inference result.
  • the design of AI models mainly includes data collection (such as collecting training data and/or inference data), model training, and model inference. It can also include the application of inference results.
  • FIG5 shows an AI application framework
  • the data source is used to provide training data sets and inference data.
  • the AI model is obtained by analyzing or training the training data provided by the data source. Among them, the AI model represents the mapping relationship between the input and output of the model. Learning the AI model through the model training node is equivalent to using the training data to learn the mapping relationship between the input and output of the model.
  • the AI model trained in the model training link is used to perform inference based on the inference data provided by the data source to obtain the inference result.
  • This link can also be understood as: inputting the inference data into the AI model, and obtaining the output through the AI model, which is the inference result.
  • the inference result can indicate: the configuration parameters used (executed) by the execution object, and/or the operation performed by the execution object.
  • the inference result is published in the inference result application link.
  • the inference result can be uniformly planned by the execution (actor) entity, for example, the execution entity can send the inference result to one or more execution objects (for example, network devices or terminal devices, etc.) for execution.
  • the execution entity can also feedback the performance of the model to the data source to facilitate the subsequent implementation of the model update training.
  • RI is used to indicate the number of downlink transmission layers recommended by the terminal device
  • CQI is used to indicate the modulation and coding mode supported by the current channel conditions determined by the terminal device
  • PMI is used to indicate the precoding recommended by the terminal device.
  • the number of precoding layers indicated by PMI corresponds to RI.
  • the introduction of AI technology into wireless communication networks has resulted in a CSI feedback method based on the AI model.
  • the terminal device uses the AI model to compress and feedback the CSI
  • the network device uses the AI model to restore the compressed CSI.
  • a sequence (such as a bit sequence) is transmitted, and the overhead is lower than that of traditional CSI feedback.
  • the encoder in FIG. 4 may be a CSI generator, and the decoder may be a CSI reconstructor.
  • the encoder may be deployed in a terminal device, and the decoder may be deployed in a network device.
  • the terminal device may generate CSI feedback information z from the CSI original information V through the encoder.
  • the terminal device reports a CSI report, which may include the CSI feedback information z.
  • the network device may reconstruct the CSI information through the decoder, that is, obtain CSI recovery information V'.
  • the training data used to train the AI model includes training samples and sample labels.
  • the training samples are channel information determined by the terminal device, and the sample labels are real channel information, i.e., true CSI.
  • the training data may only include training samples, or the training samples are sample labels.
  • the true CSI may be a high-precision CSI.
  • network element A sends information A to network element B
  • network element B in each embodiment of the present application can be understood as the destination end of the information A or the intermediate network element in the transmission path between the destination end and the network element B, which may include directly or indirectly sending information to network element B.
  • Network element B receives information A from network element A can be understood as the source end of the information A or the intermediate network element in the transmission path between the source end and the network element A, which may include directly or indirectly receiving information from network element A.
  • the information may be processed as necessary between the source end and the destination end of the information transmission, such as format changes, etc., but the destination end can understand the valid information from the source end. Similar expressions in the present application can be understood similarly and will not be repeated here.
  • FIG. 6 it is a flow chart of a model data acquisition method provided by an embodiment of the present application.
  • the method can be applied to the aforementioned model data acquisition system, such as the model data acquisition system shown in Figure 1.
  • the example shown in Figure 6 is an example of the execution subject of the interaction diagram using the first node (data synthesis node, or it can be called a central node, a first entity, etc.) and the second node (data providing node, or it can be called a node to be served, a second entity, etc.) as an example.
  • It can further include model processing (such as model training or updating, etc.) nodes and model use (such as model reasoning, etc.) nodes.
  • the model processing node can be the first node, or it can be other nodes.
  • the model use node can be the second node, or it can be other nodes, etc.
  • the model data acquisition method shown in Figure 6 may include steps 601-602. Steps 601-602 are as follows:
  • a second node sends first information to a first node, where the first information includes the following types of information: original data and characteristic information of the original data.
  • the first node receives the first information.
  • the first node may be a data synthesis node, or may be referred to as a data generation node, a central node, a first entity, etc.
  • the second node may be a data providing node, or may be referred to as a node to be served, a second entity, etc.
  • the first node is an access network device, and the second node is a terminal device.
  • Data labeling (or data annotation) is part of the preprocessing phase when developing a machine learning (ML) model. It is responsible for identifying raw data (such as images, text files, videos), and then one or more data labels can be added to the raw data to specify the context of the model and help the machine learning model make accurate predictions. Data labeling supports a variety of different machine learning and deep learning use cases, including computer vision and natural language processing (NLP).
  • NLP natural language processing
  • the priority of the original data can be taken from the predefined P level.
  • the meaning of each priority level can be predefined.
  • the generation model For example, for the original data with priority level P1, the generation model generates N1 synthetic data similar to it, and for the original data with priority level P2, the generation model generates N2 synthetic data similar to it.
  • the number of synthetic data generated by the generation model that is similar to the original data with priority level P1 is x1
  • the number of synthetic data generated by the generation model that is similar to the original data with priority level P2 is x2
  • the ratio of x1 to x2 is equal to the ratio of N1 to N2, where N1 and N2 are predefined values.
  • the way in which the second node obtains the first information also includes a way to obtain the data label of the original data.
  • the second information indicates the way or standard by which the second node determines the data label of the data.
  • E1 indicates that the data label is determined according to sparsity (for example, according to a threshold of the sparsity of the channel, the channel label is determined to be LOS or NLOS); or, E2: determines the data label according to the delay distribution, etc.
  • E4 or indicates that the data label is the RI of the channel, etc.
  • the second information may include an acquisition method identifier (for example, E1/E2/E3/E4).
  • the method may further include step 605, wherein the first node further sends fourth information to the second node, wherein the fourth information indicates that the first node has the ability to jointly process feature information of multiple original data of the same group.
  • the joint processing may, for example, be weighted summation of feature information of multiple original data of the same group.
  • the first node sends the fourth information to the second node so that the second node can divide the original data that can be jointly processed into the same group, while dividing the original data that cannot be jointly processed into the same group, thereby avoiding the fusion of original data from different groups to generate unexpected synthetic data.
  • data generation is performed with a small amount of data, which can simultaneously meet the requirements of high diversity of the synthesized data and close feature distribution of the synthesized data to the feature distribution of the original data.
  • the original data is the channel information
  • the characteristic information of the original data may include channel angle power spectrum characteristics, channel delay spectrum characteristics, etc. Then the generation model can know that the generated channel should be similar to the original channel information in angle and delay power spectrum distribution, while other characteristic values, such as the randomness of Doppler frequency deviation, can be higher.
  • weighted summing of feature information of multiple original data of the same group can improve the diversity of synthesized data.
  • the original data is channel information
  • the feature information of the original data may include channel angle power spectrum features, channel delay spectrum features, etc.
  • the above-mentioned multiple synthetic data can be used as the input of the model, and the input can be used for model processing, such as model training, model updating, model reasoning or model performance monitoring.
  • the model training, updating, etc. can be performed at the first node, or the synthesized data can be transmitted to the model processing node, etc., for model training, updating, etc.
  • the model can be inferred, or the model can be transmitted to the model using node for model inference. For example, it can be transmitted to the second node for model inference.
  • the first node further sends at least a subset of the plurality of synthesized data to the second node.
  • At least one subset of the multiple synthetic data may be, for example, a subset of the synthetic data obtained by randomly sampling multiple samples from the multiple synthetic data.
  • multiple samples are randomly sampled from the synthetic data obtained according to at least one of the multiple data labels to obtain at least one subset of the synthetic data.
  • the data label of the original data has three possible values (for example, data label a, data label b and data label c), one subset of the synthetic data is a subset obtained by randomly sampling multiple samples from the synthetic data obtained by inputting data label a as a condition, another subset of the synthetic data is a subset obtained by randomly sampling multiple samples from the synthetic data obtained by inputting data label b (or data label c) as a condition, and so on.
  • a subset of the synthetic data may be obtained by randomly sampling multiple samples S1 from the synthetic data obtained by inputting data label a as a condition, randomly sampling multiple samples S2 from the synthetic data obtained by inputting data label b as a condition, and randomly sampling multiple samples S3 from the synthetic data obtained by inputting data label c as a condition, and then obtaining one of the subsets based on the multiple samples S1, multiple samples S2, and multiple samples S3, and obtaining another subset by repeating the sampling method again.
  • the subset may also be obtained in other ways, and this solution does not limit this.
  • the second node sends fifth information to the first node based on the received at least one subset.
  • the fifth information indicates an evaluation result of at least one synthetic data in the at least one subset.
  • the evaluation result indicates at least one synthetic data that needs to be eliminated in the at least one subset, or indicates at least one synthetic data that does not need to be eliminated in the at least one subset. Accordingly, the first node receives the fifth information.
  • the synthetic data to be eliminated may be, for example, synthetic data having a different characteristic distribution from the original data of the second node.
  • the second node obtains the evaluation result by determining whether the received synthetic data is similar to the actual measured data. For example, the second node calculates the distance between a synthetic data in at least one subset and its local data. If there is a local data such that the distance between the synthetic data and the local data satisfies the first condition, the synthetic data is determined to have the same characteristic distribution as the original data, i.e., it is not eliminated; if the distance between the synthetic data and all local data does not meet the first condition, the synthetic data is determined to have a different characteristic distribution from the original data, i.e., it needs to be eliminated.
  • the first condition may be a certain specific range, or a certain threshold, etc.
  • the calculation of the distance between the synthetic data and the local data may be to use the norm of the difference between the synthetic data and the local data as the distance between the synthetic data and the local data.
  • the data feature information of the synthetic data and the local data is extracted respectively, and the norm of the difference between the data feature information of the synthetic data and the local data is used as the distance between the synthetic data and the local data.
  • the first node filters the multiple synthetic data based on the received evaluation result, and thereby obtains at least one synthetic data after filtering.
  • the at least one synthesized data after screening is determined based on the distance between the first synthesized data to be eliminated and other synthesized data in the plurality of synthesized data.
  • the first synthesized data to be eliminated is the synthesized data to be eliminated determined according to the fifth information.
  • the first node calculates the distance between the synthetic data to be removed and other synthetic data in the first node. If there is a synthetic data such that the distance between the synthetic data and the synthetic data to be removed is not greater than a certain threshold, it is determined that the synthetic data is also removed.
  • the at least one synthetic data after screening is used as the input of the model.
  • step 601 For the introduction of this part, please refer to the description of step 601, which will not be repeated here.
  • the second node is a third-party device that performs the aforementioned second node-related actions.
  • the above step 601 is performed by a third-party device.
  • the first node is a network device.
  • the above steps 601 and 602 are both performed by the network device.
  • the second node is a terminal device.
  • the above step 601 is performed by the terminal device.
  • the first node includes a network device and a third-party device.
  • the above step 602 can be performed by a third-party device, such as an OTT, or a cloud server, and the above step 601 can be performed by the network device.
  • the network device and the third-party device can also communicate with each other to transmit the content transmitted in the above step 601.
  • the second information indicates the method for obtaining the characteristic information of the original data as an example.
  • the second information can also be sent by the third node to the second node (the embodiment shown in Figure 7). That is to say, the third node indicates the method for obtaining the characteristic information of the original data to the second node, so that the second node processes based on the method for obtaining the characteristic information of the original data to obtain the characteristic information of the original data.
  • the third node can be a model training node or a model use node, and can also be called a third entity, such as a model training entity or a model use entity.
  • the first node may process the original data to obtain the characteristic information of the original data (the embodiment shown in FIG8 and the embodiment shown in FIG9 ) or the third node may process the original data to obtain the characteristic information of the original data (the embodiment shown in FIG10 ).
  • FIG. 7 it is a flow chart of another model data acquisition method provided by an embodiment of the present application.
  • the example shown in Figure 7 is illustrated by taking the first node (data synthesis node, or it can be called a central node, a first entity, etc.), the second node (data providing node, or it can be called a node to be served, a second entity, etc.) and the third node (model training node or model use node, or it can be called a third entity, such as a model training entity or a model use entity) as an example of the execution subject of the interaction diagram.
  • the difference between this example and the example shown in Figure 6 is that the second information is sent by the third node to the second node, rather than by the first node to the second node.
  • the model data acquisition method shown in Figure 7 may include steps 701-705. Steps 701-705 are as follows:
  • a first node sends third information to a second node, where the third information indicates a reporting configuration of the first information.
  • the second node receives the third information.
  • the second node processes the original data based on the method for obtaining the characteristic information of the original data from the third node to obtain the characteristic information of the original data. Furthermore, the second node sends the original data and the characteristic information of the original data to the first node. The first node generates multiple synthetic data based on the original data from the second node and the characteristic information of the original data.
  • the model training node or the model use node determines the method for obtaining the characteristic information of the original data (or the method for obtaining the data label of the original data) according to the model requirements, which can improve the quality of the synthetic data, thereby helping to improve the performance of model training or updating.
  • the second node sends sixth information to the first node, where the sixth information includes the following type of information: original data.
  • the first node receives the sixth information.
  • the third node sends the eighth information to the second node, the eighth information indicating the processing method of the second node to obtain the sixth information, and the processing method of the second node to obtain the sixth information includes the method of obtaining the data label of the original data.
  • This solution does not limit this.
  • the second node is a third-party device that performs the aforementioned actions related to the second node.
  • the above steps 1001-1003 are performed by a third-party device.
  • the third node is a third-party device that performs the aforementioned actions related to the third node.
  • the above steps 1003-1006 are performed by a third-party device.
  • the first node is a network device.
  • the above steps 1001, 1002, 1005-1007 are all performed by the network device.
  • the second node is a terminal device.
  • the above steps 1001-1003 are performed by the terminal device.
  • the third node is an AI node.
  • the above steps 1003-1006 are performed by the AI node.
  • the first node includes a network device and a third-party device.
  • the above step 1007 can be performed by a third-party device, such as an OTT, or a cloud server, and the above steps 1001, 1002, 1005, and 1006 can be performed by the network device.
  • the network device and the third-party device can also communicate with each other to transmit the content transmitted in the above steps 1001, 1002, 1005, and 1006.
  • the third node includes a third-party device, and a network device or a terminal device.
  • the above step 1004 can be performed by a third-party device, such as an OTT, or a cloud server, and the above steps 1003, 1005, and 1006 can be performed by a network device or a terminal device.
  • the network device or the terminal device can also communicate with the third-party device to transmit the content transmitted in the above steps 1003, 1005, and 1006.
  • the second node and the third node in the embodiment of the present application may be located in the same node or in two different nodes, and this solution does not impose any restrictions on this.
  • the division of multiple units or modules is only a logical division according to function, and is not used as a limitation on the specific structure of the device.
  • some functional modules may be subdivided into more small functional modules, and some functional modules may also be combined into one functional module, but no matter whether these functional modules are subdivided or combined, the general process performed by the device is the same.
  • some devices contain a receiving unit and a sending unit.
  • the sending unit and the receiving unit can also be integrated into a communication unit, which can implement the functions implemented by the receiving unit and the sending unit.
  • each unit corresponds to its own program code (or program instruction), and when the program code corresponding to each of these units is run on the processor, the unit is controlled by the processing unit to execute the corresponding process to implement the corresponding function.
  • a model data acquisition device including a module (or means) for implementing each step performed by the first node in any of the above methods.
  • a model data acquisition device including a module (or means) for implementing each step performed by the second node in any of the above methods.
  • a model data acquisition device is provided, including a module (or means) for implementing each step performed by the third node in any of the above methods.
  • the memory 1203 may be located in the one or more processors, or located outside the one or more processors, or may include a storage part located in the one or more processors and a storage part located outside the one or more processors.

Landscapes

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

Abstract

La présente demande concerne un procédé, un appareil et un système d'acquisition de données d'un modèle d'intelligence artificielle (IA). Le procédé peut comprendre les étapes consistant à : recevoir des premières informations provenant d'un second nœud, les premières informations contenant les types d'informations suivants, à savoir des données d'origine et des informations sur les caractéristiques des données d'origine ; et, sur la base des premières informations, générer de multiples données synthétiques, les multiples données synthétiques étant utilisées pour un traitement de modèle. Dans ledit exemple, des données synthétiques qui satisfont des exigences de diversité et ont la même distribution de données que des données mesurées peuvent être générées sur la base de données d'origine et d'informations sur les caractéristiques des données d'origine. Par conséquent, la qualité des données synthétiques est améliorée, ce qui améliore les performances d'un entraînement ou d'une mise à jour d'un modèle.
PCT/CN2024/143477 2023-12-29 2024-12-28 Procédé, appareil et système d'acquisition de données d'un modèle Pending WO2025140663A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202311871100.7 2023-12-29
CN202311871100.7A CN120234608A (zh) 2023-12-29 2023-12-29 模型数据获取方法及装置、系统

Publications (1)

Publication Number Publication Date
WO2025140663A1 true WO2025140663A1 (fr) 2025-07-03

Family

ID=96159352

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2024/143477 Pending WO2025140663A1 (fr) 2023-12-29 2024-12-28 Procédé, appareil et système d'acquisition de données d'un modèle

Country Status (2)

Country Link
CN (1) CN120234608A (fr)
WO (1) WO2025140663A1 (fr)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113159145A (zh) * 2018-04-28 2021-07-23 华为技术有限公司 一种特征工程编排方法及装置
CN114548416A (zh) * 2020-11-26 2022-05-27 华为技术有限公司 数据模型训练方法及装置
CN115423031A (zh) * 2022-09-20 2022-12-02 腾讯科技（深圳）有限公司 一种模型训练的方法以及相关装置
CN116318481A (zh) * 2021-12-20 2023-06-23 华为技术有限公司 一种通信方法及装置
CN116665219A (zh) * 2023-05-26 2023-08-29 华为技术有限公司 一种数据处理方法及其装置

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113159145A (zh) * 2018-04-28 2021-07-23 华为技术有限公司 一种特征工程编排方法及装置
CN114548416A (zh) * 2020-11-26 2022-05-27 华为技术有限公司 数据模型训练方法及装置
US20230281513A1 (en) * 2020-11-26 2023-09-07 Huawei Technologies Co., Ltd. Data model training method and apparatus
CN116318481A (zh) * 2021-12-20 2023-06-23 华为技术有限公司 一种通信方法及装置
CN115423031A (zh) * 2022-09-20 2022-12-02 腾讯科技（深圳）有限公司 一种模型训练的方法以及相关装置
CN116665219A (zh) * 2023-05-26 2023-08-29 华为技术有限公司 一种数据处理方法及其装置

Also Published As

Publication number Publication date
CN120234608A (zh) 2025-07-01

Similar Documents

Publication Publication Date Title
US20240354591A1 (en) Communication method and apparatus
US20250141525A1 (en) Communication method and apparatus
US20250211311A1 (en) Beam management method
WO2023115254A1 (fr) Procédé et dispositif de traitement de données
KR20250125424A (ko) Ai 모델의 훈련 데이터를 전송하는 방법 및 통신 장치
WO2025140663A1 (fr) Procédé, appareil et système d'acquisition de données d'un modèle
CN118509014A (zh) 一种通信的方法和通信装置
EP4661464A1 (fr) Procédé et appareil de communication
WO2025185425A1 (fr) Modèle sans fil, procédé et dispositif de traitement d'informations, et système
WO2025161853A1 (fr) Procédé, appareil et système de surveillance de modèle
US20250202560A1 (en) Communication method and apparatus
CN120786657A (zh) 信息传输方法、装置及系统
WO2025161598A1 (fr) Procédé d'entraînement de réseau et appareil de communication
CN120786656A (zh) 信息传输方法、装置及系统
WO2025067480A1 (fr) Procédé, appareil et système de communication
WO2025140003A1 (fr) Procédé de communication et appareil de communication
WO2024208316A1 (fr) Procédé de rétroaction d'informations de compression et appareil
WO2025232813A1 (fr) Procédé de communication et appareil de communication
WO2025209114A1 (fr) Procédé et appareil d'identification de service, et système
WO2025209114A9 (fr) Procédé et appareil d'identification de service, et système
CN120474649A (zh) 通信的方法和通信装置
CN121013063A (zh) 一种用于决策的通信方法及通信装置
CN120934717A (zh) 一种人工智能模型监控的方法和通信装置
CN120835309A (zh) 通信方法、装置、芯片模组、存储介质及程序产品
CN120835314A (zh) 一种通信方法及装置

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: 24911577

Country of ref document: EP

Kind code of ref document: A1