WO2024125421A1 - 一种模型同步方法、装置、设备及存储介质 - Google Patents

一种模型同步方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2024125421A1
WO2024125421A1 PCT/CN2023/137609 CN2023137609W WO2024125421A1 WO 2024125421 A1 WO2024125421 A1 WO 2024125421A1 CN 2023137609 W CN2023137609 W CN 2023137609W WO 2024125421 A1 WO2024125421 A1 WO 2024125421A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
time
model adjustment
adjustment
adjustment information
Prior art date
Application number
PCT/CN2023/137609
Other languages
English (en)
French (fr)
Inventor
彦楠
王达
曾二林
梁靖
Original Assignee
大唐移动通信设备有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 大唐移动通信设备有限公司 filed Critical 大唐移动通信设备有限公司
Publication of WO2024125421A1 publication Critical patent/WO2024125421A1/zh

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00Machine learning

Definitions

  • the present disclosure relates to the field of communication technology, and in particular to a model synchronization method, device, equipment and storage medium.
  • AI artificial intelligence
  • ML machine learning
  • AI models can be used to improve the performance of communication systems.
  • AI models may have unilateral models or bilateral models.
  • the bilateral model means that the AI model is divided into part A and part B, and part A and part B are deployed on two devices respectively. Part A and part B need to be executed together to perform the correct derivation process. However, it is currently impossible to adjust part A and part B synchronously, which may cause the models used by the two devices at the same time to mismatch, resulting in derivation failure.
  • the present disclosure relates to a model synchronization method, device, equipment and storage medium.
  • an embodiment of the present disclosure provides a model synchronization method, which is applied to a first device in a communication system, wherein the communication system further includes a second device, and the method includes:
  • the first AI model is adjusted to the second AI model.
  • determining the model adjustment time includes:
  • the model adjustment time is determined according to the model adjustment information.
  • the model adjustment information includes: a model synchronization period and a model adjustment duration
  • the step of determining the model adjustment time according to the model adjustment information includes:
  • the model adjustment time is determined according to the model synchronization period and the model adjustment duration.
  • determining the model adjustment time according to the model synchronization period and the model adjustment duration includes:
  • the model adjustment time is determined according to the indication information, the model synchronization period and the model adjustment duration.
  • the model adjustment information includes a preset time and an offset duration
  • the step of determining the model adjustment time according to the model adjustment information includes:
  • the model adjustment time is determined according to the preset time and the offset duration.
  • the model adjustment information further includes: a reference time mode and/or a model synchronization time accuracy.
  • the communication system further includes a third-party management device.
  • the model adjustment information is configured through the first device, the second device, or the third-party management device;
  • the model adjustment information is determined through negotiation between the first device and the second device;
  • the model adjustment information is determined through negotiation between the third-party management device and the first device;
  • the model adjustment information is determined through negotiation between the third-party management device and the second device;
  • the model adjustment information is determined through negotiation among the third-party management device, the first device, and the second device.
  • adjusting the first AI model to the second AI model includes any one of the following:
  • the method further comprises:
  • the first AI model or the second AI model is determined as the target AI model for processing the first data
  • the initial transmission time is the initial transmission time of the second data
  • the second data is the data obtained after the target AI model processes the first data.
  • determining the first AI model or the second AI model as a target AI model for processing the first data according to the initial transmission time and the model adjustment time includes:
  • the second AI model is determined as the target AI model.
  • the first device includes a terminal device or a network device; the second device includes a terminal device or a network device.
  • the method further comprises:
  • the second AI model sent by the second device or the third-party management device is received, where the second AI model is received before the model adjustment time or before the determination time of the model adjustment time.
  • the method further comprises:
  • an embodiment of the present disclosure provides a first device, which is applied to a communication system. Also included is a second device, wherein the first device includes a memory, a transceiver, and a processor:
  • the memory is used to store computer programs
  • the transceiver is used to send and receive data under the control of the processor
  • the processor is configured to read the computer program in the memory and perform the following operations:
  • the first AI model is adjusted to the second AI model.
  • the processor is specifically configured to perform the following operations:
  • the model adjustment time is determined according to the model adjustment information.
  • the model adjustment information includes: a model synchronization period and a model adjustment duration
  • the processor is specifically configured to perform the following operations:
  • the model adjustment time is determined according to the model synchronization period and the model adjustment duration.
  • the processor is specifically configured to perform the following operations:
  • the model adjustment time is determined according to the indication information, the model synchronization period and the model adjustment duration.
  • the model adjustment information includes a preset time and an offset duration
  • the processor is specifically configured to perform the following operations:
  • the model adjustment time is determined according to the preset time and the offset duration.
  • the model adjustment information further includes: a reference time mode and/or a model synchronization time accuracy.
  • the communication system further includes a third-party management device.
  • the model adjustment information is configured through the first device, the second device, or the third-party management device;
  • the model adjustment information is determined through negotiation between the first device and the second device;
  • the model adjustment information is determined through negotiation between the third-party management device and the first device;
  • the model adjustment information is determined through negotiation between the third-party management device and the second device;
  • the model adjustment information is determined through negotiation among the third-party management device, the first device, and the second device.
  • the processor is specifically configured to perform any one of the following operations:
  • the processor is further configured to perform the following operations:
  • the first AI model or the second AI model is determined as the target AI model for processing the first data
  • the initial transmission time is the initial transmission time of the second data
  • the second data is the data obtained after the target AI model processes the first data.
  • the processor is specifically configured to perform the following operations:
  • the second AI model is determined as the target AI model.
  • the processor is further configured to perform the following operations:
  • the second AI model sent by the second device or the third-party management device is received, where the second AI model is received before the model adjustment time or before the determination time of the model adjustment time.
  • the processor is further configured to perform the following operations:
  • an embodiment of the present disclosure provides a model synchronization device, including:
  • a first determining unit configured to determine a model adjustment time at which the first device and the second device synchronously perform model adjustment
  • An adjustment unit is used to adjust the first AI model to the second AI model according to the model adjustment time.
  • the first determining unit is specifically configured to:
  • the model adjustment time is determined according to the model adjustment information.
  • the model adjustment information includes: a model synchronization period and a model adjustment duration; and the first determining unit is specifically configured to:
  • the model adjustment time is determined according to the model synchronization period and the model adjustment duration.
  • the first determining unit is specifically configured to:
  • the model adjustment time is determined according to the indication information, the model synchronization period and the model adjustment duration.
  • the model adjustment information includes a preset time and an offset duration; and the first determining unit is specifically configured to:
  • the model adjustment time is determined according to the preset time and the offset duration.
  • the model adjustment information further includes: a reference time mode and/or a model synchronization time accuracy.
  • the communication system further includes a third-party management device.
  • the model adjustment information is configured through the first device, the second device, or the third-party management device;
  • the model adjustment information is determined through negotiation between the first device and the second device;
  • the model adjustment information is determined through negotiation between the third-party management device and the first device;
  • the model adjustment information is determined through negotiation between the third-party management device and the second device;
  • the model adjustment information is determined through negotiation among the third-party management device, the first device, and the second device.
  • the adjustment unit is specifically configured to implement any one of the following:
  • the device further includes a second determining unit, wherein the second determining unit is configured to:
  • the first AI model or the second AI model is determined as the target AI model for processing the first data
  • the initial transmission time is the initial transmission time of the second data
  • the second data is the data obtained after the target AI model processes the first data.
  • the second determining unit is specifically configured to:
  • the second AI model is determined as the target AI model.
  • the device further includes a receiving unit, wherein the receiving unit is configured to:
  • the second AI model sent by the second device or the third-party management device is received, where the second AI model is received before the model adjustment time or before the determination time of the model adjustment time.
  • the device further includes a sending unit, wherein the sending unit is configured to:
  • an embodiment of the present disclosure provides a processor-readable storage medium, wherein the processor-readable storage medium stores a computer program, and the computer program is used to enable the processor to execute the method described in the first aspect.
  • the present disclosure provides a model synchronization method, apparatus, device and storage medium, in which all devices using the same AI model first determine the model adjustment time, and each device using the same AI model then adjusts the time according to the model, and adjusts the first AI model to the second AI model at the same time. Multiple devices using the same AI model can adjust the model at the same time, so that the AI models used by multiple devices using the same AI model at the same time can be completely matched.
  • FIG1 is a schematic diagram of an architecture of a communication system provided by an embodiment of the present disclosure.
  • FIG2 is a flow chart of a model synchronization method provided by an embodiment of the present disclosure
  • FIG3 is a flowchart 1 of the negotiation model adjustment information provided by an embodiment of the present disclosure.
  • FIG4 is a second flowchart of the negotiation model adjustment information provided by an embodiment of the present disclosure.
  • FIG5 is a schematic diagram of determining a model adjustment time provided by an embodiment of the present disclosure.
  • FIG6 is a schematic diagram of another model synchronization method provided by an embodiment of the present disclosure.
  • FIG7 is a schematic diagram of determining a target AI model provided by an embodiment of the present disclosure.
  • FIG8A is a first structural diagram of a first device provided in an embodiment of the present disclosure.
  • FIG8B is a second structural diagram of the first device provided in an embodiment of the present disclosure.
  • FIG9A is a first structural diagram of a model synchronization device provided by an embodiment of the present disclosure.
  • FIG9B is a second structural diagram of a model synchronization device provided by an embodiment of the present disclosure.
  • FIG9C is a third structural diagram of a model synchronization device provided by an embodiment of the present disclosure.
  • FIG9D is a fourth structural diagram of a model synchronization device provided in an embodiment of the present disclosure.
  • the term "and/or” describes the association relationship of associated objects, indicating that three relationships may exist.
  • a and/or B may represent three situations: A exists alone, A and B exist at the same time, and B exists alone.
  • the character "/" generally indicates that the associated objects before and after are in an "or” relationship.
  • plurality in the embodiments of the present disclosure refers to two or more than two, and other quantifiers are similar thereto.
  • the embodiments of the present disclosure provide a model synchronization method, whereby multiple devices using an AI model can adjust the models at the same time, thereby making the AI models used by the multiple devices using the AI model at the same time completely match.
  • the method and the device are based on the same application concept. Since the method and the device solve the problem in a similar principle, the implementation of the device and the method can refer to each other, and the repeated parts will not be repeated.
  • the applicable systems can be global system of mobile communication (GSM) system, code division multiple access (CDMA) system, wideband code division multiple access (WCDMA) general packet radio service (GPRS) system, long term evolution (LTE) system, LTE frequency division duplex (FDD) system, LTE time division duplex (TDD) system, etc. (time division duplex, TDD) system, long term evolution advanced (long term evolution advanced, LTE-A) system, universal mobile telecommunication system (UMTS), worldwide interoperability for microwave access (WiMAX) system, 5G new radio (NR) system, etc.
  • GSM global system of mobile communication
  • CDMA code division multiple access
  • WCDMA wideband code division multiple access
  • GPRS general packet radio service
  • LTE long term evolution
  • FDD frequency division duplex
  • TDD LTE time division duplex
  • UMTS universal mobile telecommunication system
  • WiMAX worldwide interoperability for microwave access
  • NR new radio
  • the terminal device involved in the embodiments of the present disclosure may be a device that provides voice and/or data connectivity to a user, a handheld device with a wireless connection function, or other processing devices connected to a wireless modem.
  • the names of terminal devices may also be different.
  • the terminal device may be called a user equipment (UE).
  • UE user equipment
  • a wireless terminal device may communicate with one or more core networks (CN) via a radio access network (RAN).
  • CN core networks
  • RAN radio access network
  • the wireless terminal device may be a mobile terminal device, such as a mobile phone (or a "cellular" phone) and a computer with a mobile terminal device.
  • the wireless terminal device may also be referred to as a system, a subscriber unit, a subscriber station, a mobile station, a mobile station, a remote station, an access point, a remote terminal device, an access terminal device, a user terminal device, a user agent, and a user device, but is not limited to these in the embodiments of the present disclosure.
  • the network device involved in the embodiments of the present disclosure may be an access network device or a core network device.
  • the access network equipment can be a base station, which can include multiple cells that provide services to the terminal.
  • the base station can also be called an access point, or it can be a device in the access network that communicates with the wireless terminal device through one or more sectors on the air interface, or other names.
  • the network device can be used to interchange received air frames with Internet Protocol (IP) packets, and act as a router between the wireless terminal device and the rest of the access network, where the rest of the access network may include an Internet Protocol (IP) communication network.
  • IP Internet Protocol
  • the network device can also coordinate the attribute management of the air interface.
  • the network device involved in the embodiments of the present disclosure may be a network device (base transceiver station, BTS) in the global system for mobile communications (GSM) or code division multiple access (CDMA), or a network device (NodeB) in wide-band code division multiple access (WCDMA), or an evolved network device (evolutional Node B, eNB or e-NodeB) in the long term evolution (LTE) system, a 5G base station (gNB) in the 5G network architecture (next generation system), or a home evolved Node B (HeNB), a relay node, a home base station (femto), a pico base station (pico), etc., but is not limited in the embodiments of the present disclosure.
  • network devices may include centralized unit (CU) nodes and distributed unit (DU) nodes, and the centralized unit and the distributed unit may also be geographically separated.
  • the core network equipment may include any of the following: a network data analytics function (NWDAF) entity, a policy control function (PCF) entity, an application function ( function, AF) entity, access and mobility management function (AMF) entity, session management function (SMF) entity, network exposure function (NEF) entity, user plane function (UPF) entity, unified data repository (UDR) entity, network slice selection function (NSSF) entity, authentication server function (AUSF) entity, unified data management (UDM) entity, network function database function (NRF) entity.
  • NWDAF network data analytics function
  • PCF policy control function
  • AMF application function
  • AMF access and mobility management function
  • SMF session management function
  • NEF network exposure function
  • UPF user plane function
  • UPF unified data repository
  • NSSF network slice selection function
  • AUSF authentication server function
  • UDM unified data management
  • NRF network function database function
  • Figure 1 is a schematic diagram of an architecture of a communication system provided by an embodiment of the present disclosure.
  • the architecture includes terminal devices, network devices, and third-party management devices.
  • terminal devices, network devices, and third-party management devices can all be initiators of AI model changes, and terminal devices and network devices can all be users of AI models.
  • the present disclosure proposes the following technical concept: all devices using the same AI model first determine the model adjustment time, and each device using the same AI model then adjusts the time according to the model, and adjusts the first AI model to the second AI model at the same time. Multiple devices using the same AI model can adjust the model at the same time, so that the AI models used by multiple devices using the same AI model at the same time can be fully matched.
  • FIG2 is a flow chart of a model synchronization method provided by an embodiment of the present disclosure. As shown in FIG2 , the method includes:
  • S201 Determine a model adjustment time at which a first device and a second device synchronously perform model adjustment.
  • the first device and the second device are devices in a communication system.
  • the first device may be a terminal device or a network device.
  • the second device may be a terminal device or a network device.
  • the first device and the second device can both be initiators of model changes and users of the model.
  • the communication system may also include a third-party management device, which may serve as an initiator of model changes.
  • the model adjustment time may be determined in the following manner: obtaining model adjustment information, and determining the model adjustment time according to the model adjustment information.
  • the model adjustment information includes at least one of the following: a model synchronization period, a model adjustment duration, a preset time, an offset duration, a reference time mode, and a model synchronization time accuracy.
  • model adjustment information may include any of the following:
  • Preset time, offset duration, reference time method and model synchronization time accuracy Preset time, offset duration, reference time method and model synchronization time accuracy.
  • the model synchronization period can be pre-set by the initiator of the model change, or the initiator of the model change can monitor the performance of the system and determine the model synchronization period based on the monitored system performance results. For example, if the monitored system performance results are good, a longer model synchronization period can be set; if the monitored system performance results are poor, a shorter model synchronization period can be set.
  • the model adjustment duration may refer to the offset duration between the model adjustment moment and the start moment of the model synchronization cycle. In a possible implementation, the model adjustment duration is shorter than the duration of the model synchronization cycle.
  • the preset time may refer to an absolute time, the preset time may refer to a model adjustment time, or may refer to any absolute time.
  • the offset duration may refer to the offset duration between the model adjustment time and the preset time.
  • the reference time method refers to the reference time method for initial time synchronization of multiple model users.
  • the initial time synchronization of multiple model users can be achieved through the global positioning system (GPS) + coordinated universal time (UTC) method, or the 1588 method.
  • GPS global positioning system
  • UTC coordinated universal time
  • Model synchronization time accuracy can be used to determine the accuracy of the model adjustment moment.
  • the model synchronization time accuracy can be the model synchronization time accuracy at the slot level, the model synchronization time accuracy at the subframe level, or the model synchronization time accuracy at the frame level; different model synchronization time accuracy may be required for model deployment for different use cases, or for different AI model change operations, for example, the model synchronization time accuracy at the slot level may be required for channel state information (CSI) feedback, while the model synchronization time accuracy at the frame level may be required for positioning.
  • CSI channel state information
  • a higher model synchronization time accuracy e.g., slot level
  • a lower model synchronization time accuracy e.g., frame level
  • model adjustment information can be obtained in the following way:
  • the initiator of the model change configures the model adjustment information and indicates the model adjustment information to the model user; or, the initiator of the model change negotiates with the model user to determine the model adjustment information.
  • the first device adjusts the first AI model to the second AI model according to the model adjustment time.
  • the AI model may include an AI model and/or an ML model.
  • the first AI model may be adjusted to the second AI model in any of the following ways:
  • the second AI model may be determined by an initiator of the model change, and the initiator of the model change may send the second AI model to the model user.
  • the second AI model may be determined and sent in any of the following ways:
  • the third-party management device determines the second AI model and sends the second AI model to the first device
  • the third-party management device determines the second AI model and sends the second AI model to the second device
  • the first device determines a second AI model and sends the second AI model to the second device
  • the second device determines a second AI model and sends the second AI model to the first device.
  • the third-party management device determines the second AI model and sends the second AI model to multiple terminal devices;
  • the third-party management device determines a second AI model and sends the second AI model to multiple network devices
  • the third-party management device determines a second AI model and sends the second AI model to multiple terminal devices and network devices.
  • the refinement of the second method can refer to the refinement of the first method, which will not be repeated here.
  • the first terminal device determines a second AI model and sends the second AI model to multiple second terminal devices;
  • the terminal device determines a second AI model and sends the second AI model to multiple network devices
  • the first terminal device determines a second AI model, and sends the second AI model to multiple second terminal devices and network devices;
  • the first network device determines a second AI model and sends the second AI model to multiple second network devices
  • the network device determines a second AI model and sends the second AI model to multiple terminal devices
  • the first network device determines a second AI model and sends the second AI model to multiple second network devices and terminal devices.
  • the refinement of the fourth method can refer to the refinement of the third method, which will not be repeated here.
  • the initiator of the model change needs to determine the second AI model before the model adjustment time; the model user needs to receive the second AI model before the model adjustment time or before the determination time of the model adjustment time.
  • replacing the first AI model with the second AI model may refer to deleting the first AI model and adding the second AI model.
  • updating the first AI model to the second AI model may refer to deleting the first AI model and adding the second AI model; or, based on the first AI model, the code of the first AI model is added, deleted or modified to obtain the second AI model.
  • deactivating the first AI model may refer to terminating the activation state of the first AI model.
  • deactivating the first AI model may be achieved in the following manner:
  • the model user disables or enables the entire AI function/use case. Since the entire AI function/use case includes one or more AI models, if the entire AI function/use case is disabled, all AI models in the entire AI function/use case will be disabled; if the entire AI function/use case is enabled, all AI models in the entire AI function/use case will be enabled.
  • the second AI model is not enabled, it is possible to fall back to the traditional function, for example, CSI feedback falls back to basic codebook parameter reporting.
  • each AI use case/function in the model user may include one or more AI models, multiple AI models of the same AI use case/function may be switched, and multiple or a single model may be added, deleted, modified, replaced, updated, activated and/or deactivated.
  • the first device may adjust the first AI model to the second AI model at the model adjustment moment.
  • the format/form of the data intermediate results transmitted by different model users can also be used to determine when to adjust the model. That is, the data receiver can compare the data format/form of the data intermediate results derived by the AI model with the format/form of the data intermediate results obtained by the traditional codebook method, and then determine whether to perform model adjustment at the receiver based on the format/form of the data intermediate results of the two.
  • the second device adjusts the first AI model to the second AI model according to the model adjustment time.
  • the first AI model in the second device and the first AI model in the first device belong to the same AI model; the second AI model in the second device and the second AI model in the first device also belong to the same AI model; it’s just that each device includes different parts of the same model.
  • the complete AI model consists of part A and part B.
  • AI model part A is used for encoding derivation based on the AI model on the terminal device side
  • AI model part B is used for decoding derivation based on the AI model on the network device side.
  • Both the first device and the second device can adjust the model according to the same model adjustment moment. That is, each model user needs to adjust the model of different parts of the same AI model at the same time.
  • the model synchronization method provided by the embodiment of the present disclosure includes: the first device determines the model adjustment time at which the first device and the second device synchronously adjust the model; and according to the model adjustment time, the first AI model is adjusted to the second AI model.
  • the second device can adjust the first AI model to the second AI model according to the model adjustment time. That is, all devices using the same AI model first determine the model adjustment time, and each device using the same AI model then adjusts the model simultaneously according to the model adjustment time, so that the AI models used by multiple devices using the same AI model at the same time can be completely matched.
  • model adjustment information can be obtained in any of the following ways:
  • the third-party management device configures model adjustment information, and multiple model users receive the model adjustment information sent by the third-party management device;
  • the first device configures model adjustment information, and multiple model users receive the model adjustment information sent by the first device;
  • the second device configures the model adjustment information, and the multiple model users receive the model adjustment information sent by the second device;
  • the first device and the second device negotiate model adjustment information
  • the third-party management device and the first device negotiate model adjustment information
  • the third-party management device and the second device negotiate model adjustment information
  • the third-party management device, the first device, and the second device negotiate model adjustment information.
  • the model user can be a terminal device or a network device.
  • the model user After the model user receives the model adjustment information sent by the initiator of the model change, it can also send confirmation information to the initiator of the model change.
  • the third-party management device configures model adjustment information, and multiple terminal devices receive the model adjustment information sent by the third-party management device;
  • the third-party management device configures the model adjustment information, and multiple network devices receive the model adjustment information sent by the third-party management device;
  • the third-party management device configures the model adjustment information, and multiple terminal devices and network devices receive the model adjustment information sent by the third-party management device.
  • the first network device configures model adjustment information, and a plurality of second network devices receive the model adjustment information sent by the first network device;
  • the network device configures the model adjustment information, and the plurality of terminal devices receive the model adjustment information sent by the first network device;
  • the first network device configures model adjustment information, and multiple terminal devices and the second network device receive the model adjustment information sent by the first network device.
  • the terminal device configures the model adjustment information, and the plurality of network devices receive the model adjustment information sent by the terminal device;
  • the first terminal device configures model adjustment information, and the plurality of second terminal devices receive the model adjustment information sent by the first terminal device;
  • the first terminal device configures model adjustment information, and multiple second terminal devices and network devices receive the model adjustment information sent by the first terminal device.
  • the refinement of the third method can refer to the refinement of the second method, which will not be repeated here.
  • the first terminal device and the second terminal device negotiate model adjustment information.
  • Network equipment and terminal equipment negotiate model adjustment information
  • the first network device and the second network device negotiate model adjustment information.
  • the fifth method can be refined into any of the following methods:
  • the refinement of the sixth method can refer to the refinement of the fifth method, which will not be repeated here.
  • the seventh method can be refined into any of the following methods:
  • the third-party management device, the first terminal device and the second terminal device negotiate model adjustment information
  • the third-party management device, the first network device and the second network device negotiate model adjustment information
  • FIG3 is a flowchart of the first embodiment of the present disclosure for negotiating model adjustment information. As shown in FIG3, taking the initiator of the model change as a network device and the model user as a network device and a terminal device as an example, the method includes:
  • a network device determines first model adjustment information.
  • the first model adjustment information has the same content as the model adjustment information in the embodiment shown in FIG. 2 , except that the specific parameter values and/or types are different.
  • the network device may determine the first model adjustment information according to the performance of the system.
  • S302 The network device sends first model adjustment information to the terminal device.
  • the network device sends the first model adjustment information to the terminal device, which may also be represented as the terminal device receiving the first model adjustment information sent by the network device.
  • the terminal device may determine the first model adjustment time according to the first model adjustment information. If the terminal device determines that it cannot perform model adjustment at the first model adjustment time, S303 may be executed.
  • S303 The terminal device determines second model adjustment information.
  • the second model adjustment information has the same content as the first model adjustment information, except for specific parameter values and/or types.
  • the second model adjustment time determined according to the second model adjustment information is different from the first model adjustment time determined according to the first model adjustment information.
  • S304 The terminal device sends second model adjustment information to the network device.
  • the terminal device sends the second model adjustment information to the network device, which may also be represented as the network device receiving the second model adjustment information sent by the terminal device.
  • the network device may determine the second model adjustment time according to the second model adjustment information, and perform model adjustment together with the terminal device at the second model adjustment time.
  • the number of negotiations between the network device and the terminal device may not be limited to the number shown in the embodiment shown in FIG. 3 .
  • the network device after receiving the second model adjustment information, if the network device determines that it cannot perform the model adjustment at the second model adjustment time, it can re-determine the third model adjustment information and send it to the terminal device, and so on, until the network device and the terminal device negotiate the model adjustment information that both parties agree on.
  • the number of negotiations should be reduced as much as possible.
  • the manner in which the devices listed in 4, 5 and 6 above negotiate the model adjustment information may refer to the negotiation manner shown in the embodiment shown in FIG. 3 .
  • FIG4 is a second flowchart of the negotiation model adjustment information provided by the embodiment of the present disclosure. As shown in FIG4, the method includes include:
  • S401 A third-party management device determines first model adjustment information.
  • the first model adjustment information has the same content as the model adjustment information in the embodiment shown in FIG. 2 , except that the specific parameter values and/or types are different.
  • the third-party management device may determine the first model adjustment information according to the performance of the system.
  • the third-party management device sends first model adjustment information to the first device and the second device respectively.
  • the third-party management device sends the first model adjustment information to the first device and the second device respectively, which can also be represented as the first device receiving the first model adjustment information sent by the third-party management device; and the second device receiving the first model adjustment information sent by the third-party management device.
  • the first device and the second device may determine the first model adjustment time according to the first model adjustment information. If the first device (or the second device) determines that it cannot perform model adjustment at the first model adjustment time, S403 may be executed.
  • S403 The first device (or the second device) determines second model adjustment information.
  • the second model adjustment information has the same content as the first model adjustment information, except for specific parameter values and/or types.
  • the second model adjustment time determined according to the second model adjustment information is different from the first model adjustment time determined according to the first model adjustment information.
  • S404 The first device (or the second device) sends second model adjustment information to the third-party management device.
  • the first device (or the second device) sends the second model adjustment information to the third-party management device, which can also be represented as the third-party management device receiving the second model adjustment information sent by the first device (or the second device).
  • S405 The third-party management device sends second model adjustment information to the second device (or the first device).
  • the third-party management device sends the second model adjustment information to the second device (or the first device), which can also be represented as the second device (or the first device) receiving the second model adjustment information sent by the third-party management device.
  • the second device After receiving the second model adjustment information, the second device (or the first device) can determine the second model adjustment time according to the second model adjustment information, and perform model adjustment together with the first device (or the second device) at the second model adjustment time.
  • multiple devices using the same AI model can first perform network synchronization, and then normal data transmission can be performed. That is, effective AI model synchronization can be performed later on the basis of network synchronization.
  • the network synchronization method can refer to the methods in the prior art, which will not be described here.
  • the model adjustment time may be determined according to the model synchronization period and the model adjustment duration.
  • the start time of the model synchronization cycle can be determined, and the model adjustment time can be determined according to the start time of the model synchronization cycle and the model adjustment duration. If the start time of the model synchronization cycle is ts and the model adjustment duration is TL, the model adjustment time is ts+TL.
  • the starting time of the model synchronization cycle can be determined based on the time of receiving the second AI model; the starting time of the model synchronization cycle can also be determined based on the time of receiving the model adjustment information; the starting time of the model synchronization cycle can also be indicated by the initiator of the model change.
  • model adjustment moment can be determined in the following manner:
  • the initiator of the model change sends an indication message to the model user, and the model user determines the model adjustment time according to the indication message, the model synchronization cycle and the model adjustment duration.
  • the model user may determine the target model synchronization period according to the indication information and the model synchronization period, and determine the model adjustment time according to the start time of the target model synchronization period and the model adjustment duration.
  • the target model synchronization period may refer to a period for performing model adjustment.
  • Figure 5 is a schematic diagram of determining a model adjustment time provided by an embodiment of the present disclosure.
  • the terminal device and the network device perform network synchronization at time t0; then the terminal device receives the model adjustment information sent by the network device at time t1, and the model adjustment information includes the model adjustment period, the model adjustment duration and the model synchronization time accuracy, wherein the model adjustment period is 200ms, the model adjustment duration is 5 time slots, and the model synchronization time accuracy is at the time slot level; the terminal device determines multiple model change periods according to the model adjustment information.
  • the terminal device receives the second AI model or indication information sent by the network device at time t2, and time t2 is in the Nth model adjustment period, the first AI model can be adjusted to the second AI model in the N+1th model adjustment period, and the specific model adjustment time can be confirmed according to the start time and model adjustment duration of the N+1th model adjustment period, wherein N>0.
  • the model adjustment duration may be 0 or greater than 0. If the model adjustment duration is 0, it means that the model adjustment is performed at the boundary of the model synchronization period; if the model adjustment duration is greater than 0, it means that the model adjustment is performed during the model adjustment period.
  • FIG6 is a schematic diagram of another model synchronization method provided by an embodiment of the present disclosure. As shown in FIG6 , taking a network device as the initiator of a model change, and taking a terminal device and a network device as a model usage room as an example, the method includes:
  • the network device and the terminal device perform network synchronization.
  • S602 The network device and the terminal device negotiate model adjustment information.
  • the model adjustment information includes the model synchronization period and the model adjustment duration.
  • the network device sends a second AI model to the terminal device.
  • S604 The network device sends instruction information to the terminal device.
  • the indication message is used to indicate that model adjustment can be started.
  • S605 The network device and the terminal device determine the model adjustment time according to the model adjustment information and the instruction information, and perform model adjustment at the model adjustment time.
  • the method of determining the model adjustment time can refer to the above embodiment and will not be repeated here.
  • the model user may determine the model adjustment time according to a preset time and an offset duration.
  • the offset duration may be 0 or greater than 0.
  • the offset duration When the offset duration is 0, it means that the preset moment can be directly used as the model adjustment moment; when the offset duration is greater than 0, it means that the moment offset from the preset moment is used as the model adjustment moment.
  • the model adjustment time is t+T.
  • the model user can determine the target AI model for processing the first data based on the initial transmission time and the model adjustment time.
  • the initial transmission time is the initial transmission time of the second data
  • the second data is the data obtained after the target AI model processes the first data.
  • the model user may determine the first AI model or the second AI model as the target AI model for processing the first data based on the initial transmission time and the model adjustment time: if the initial transmission time is before the model adjustment time, the first AI model is determined as the target AI model; if the initial transmission time is after the model adjustment time, the second AI model is determined as the target AI model.
  • the model user uses the first AI model as the target AI model regardless of whether the second data is the initial transmission or the retransmission; if the initial transmission time is after the model adjustment time, the model user uses the second AI model as the target AI model regardless of whether the second data is the initial transmission or the retransmission.
  • the second data is transmitted from the terminal device to the network device using the physical uplink control channel (PUCCH).
  • PUCCH physical uplink control channel
  • the second data is transmitted from the terminal device to the network device using the physical uplink shared channel (PUSCH)
  • PUSCH physical uplink shared channel
  • the terminal device performs an encoding operation derived based on the AI model, and sends the derived data intermediate result (second data) to the network device, and the network device performs a decoding operation derived based on the AI model:
  • the receiver When the receiver receives the second data, it is necessary to derive the AI model used by the sender in encoding.
  • the network device side can know the initial transmission time) to determine the AI model used for encoding on the terminal device side; that is, if the derived initial transmission time is before the model adjustment time, even if the model adjustment time has been reached and the old AI model (first AI model) is adjusted to the new AI model (second AI model), the old AI model must be used for decoding; therefore, it is necessary
  • the network device temporarily saves the information of the previous AI model (old AI model) used.
  • the terminal device uses the old AI model to perform derivation on the first data to obtain the second data, and transmits the second data to the network device for the first time at time t1.
  • the network device decodes the first transmitted second data at time t1, and the decoding fails;
  • time t2 is the model adjustment time, and the network device and the terminal device both adjust the old AI model to the new AI model at time t2, and the network device also needs to save the old AI model;
  • the terminal device repeats the transmission of the second data to the network device at time t3, and the network device uses the old AI model to perform derivation on the retransmitted second data at time t3.
  • the sender uses the AI model to derive the second data to be sent, if it is determined that the initial transmission should be initiated after the model adjustment moment, even if the encoding occurs before the model adjustment moment, the AI model derivation operation for encoding is required according to the subsequent new AI model.
  • Accurately determining the target AI model can ensure that the models used by each model user are fully matched.
  • FIG8A is a structural schematic diagram 1 of a first device provided by an embodiment of the present disclosure.
  • the first device includes: a memory 801 , a transceiver 802 , and a processor 803 .
  • Memory 801 used for storing computer programs
  • the transceiver 802 is used to send and receive data under the control of the processor 803;
  • the processor 803 is configured to read the computer program stored in the memory 801 and perform the following operations:
  • the first AI model is adjusted to the second AI model.
  • the processor 803 is specifically configured to perform the following operations:
  • the model adjustment time is determined according to the model adjustment information.
  • the model adjustment information includes: a model synchronization period and a model adjustment duration
  • the processor 803 is specifically configured to perform the following operations:
  • the processor 803 is specifically configured to perform the following operations:
  • the model adjustment information includes a preset time and an offset duration
  • the processor 803 is specifically configured to perform the following operations:
  • the model adjustment information further includes: a reference time mode and/or a model synchronization time accuracy.
  • the communication system further includes a third-party management device.
  • the model adjustment information is configured through the first device, the second device, or a third-party management device;
  • the model adjustment information is determined through negotiation between the first device and the second device;
  • the model adjustment information is determined through negotiation between a third-party management device and the first device;
  • the model adjustment information is determined through negotiation between a third-party management device and a second device;
  • the model adjustment information is determined through negotiation between a third-party management device, the first device, and the second device.
  • the processor 803 is specifically configured to perform any one of the following operations:
  • the processor 803 is further configured to perform the following operations:
  • the target AI model for processing the first data is determined, the initial transmission time is the initial transmission time of the second data, and the second data is the data obtained after the target AI model processes the first data.
  • the processor is specifically configured to perform the following operations:
  • the first AI model is determined as the target AI model
  • the second AI model is determined as the target AI model.
  • the processor 803 is further configured to perform the following operations:
  • a second AI model sent by a second device or a third-party management device is received, where the second AI model is received before the model adjustment time or before the determination time of the model adjustment time.
  • the processor 803 is further configured to perform the following operations:
  • the second AI model is sent to the second device.
  • the bus architecture may include any number of interconnected buses and bridges, specifically one or more processors represented by processor 803 and various circuits of memory represented by memory 801 are linked together.
  • 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 802 may be a plurality of components, namely, a transmitter and a receiver, providing a unit for communicating with various other devices on a transmission medium, which transmission medium includes a wireless channel, a wired channel, an optical cable, and other transmission media.
  • the processor 803 is responsible for managing the bus architecture and general processing, and the memory 801 may store data used by the processor 803 when performing operations.
  • FIG8B is a second structural diagram of the first device provided by the embodiment of the present disclosure.
  • the device when the first device is a terminal device, the device may further include a user interface 804.
  • the user interface 804 may also be an interface capable of connecting external or internal devices.
  • the connected devices include but are not limited to a keypad, a display, a speaker, a microphone, a joystick, etc.
  • processor 803 can be a central processing unit (CPU), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or a complex programmable logic device (CPLD), and the processor can also adopt a multi-core architecture.
  • CPU central processing unit
  • ASIC application specific integrated circuit
  • FPGA field-programmable gate array
  • CPLD complex programmable logic device
  • the processor 803 is used to execute any of the methods provided by the embodiments of the present disclosure according to the obtained executable instructions by calling the computer program stored in the memory 801.
  • the processor 803 and the memory 801 may also be arranged physically separately.
  • first device provided by the present disclosure can implement all the method steps implemented by the first device in the above-mentioned method embodiment, and can achieve the same technical effect.
  • the parts and beneficial effects of this embodiment that are the same as the method embodiment will not be described in detail here.
  • FIG9A is a structural schematic diagram 1 of a model synchronization device provided by an embodiment of the present disclosure. As shown in FIG9 , the device includes:
  • a first determining unit 901 is used to determine a model adjustment time at which the first device and the second device synchronously perform model adjustment;
  • the adjustment unit 902 is used to adjust the first AI model to the second AI model according to the model adjustment time.
  • the first determining unit 901 is specifically configured to:
  • the model adjustment time is determined.
  • the model adjustment information includes: a model synchronization period and a model adjustment duration; the first determination unit 901 is specifically configured to:
  • the model adjustment time is determined according to the model synchronization period and the model adjustment duration.
  • the first determining unit 901 is specifically configured to:
  • the model adjustment information includes a preset time and an offset duration; the first determination unit 901 is specifically configured to:
  • the model adjustment information further includes: a reference time mode and/or a model synchronization time accuracy.
  • the communication system further includes a third-party management device.
  • the model adjustment information is configured through the first device, the second device, or a third-party management device;
  • the model adjustment information is determined through negotiation between the first device and the second device;
  • the model adjustment information is determined through negotiation between a third-party management device and the first device;
  • the model adjustment information is determined through negotiation between a third-party management device and a second device;
  • the model adjustment information is determined through negotiation between a third-party management device, the first device, and the second device.
  • the adjustment unit 902 is specifically configured to implement any one of the following:
  • FIG9B is a second structural diagram of a model synchronization device provided by an embodiment of the present disclosure. As shown in FIG9B , the device further includes a second determination unit 903, and the second determination unit 903 is used to:
  • the target AI model for processing the first data is determined.
  • the initial transmission time is the initial transmission time of the second data
  • the second data is the data obtained after the target AI model processes the first data.
  • the second determining unit 903 is specifically configured to:
  • the first AI model is determined as the target AI model
  • the second AI model is determined as the target AI model.
  • FIG9C is a third structural diagram of a model synchronization device provided by an embodiment of the present disclosure. As shown in FIG9C , the device further includes a receiving unit 904, and the receiving unit 904 is used to:
  • a second AI model sent by a second device or a third-party management device is received, where the second AI model is received before the model adjustment time or before the determination time of the model adjustment time.
  • FIG9D is a fourth structural diagram of a model synchronization device provided by an embodiment of the present disclosure. As shown in FIG9D , the device further includes a sending unit 905, and the sending unit 905 is used to:
  • the second AI model is sent to the second device.
  • each functional unit in each embodiment of the present disclosure may be integrated into a processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
  • the above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
  • the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a processor-readable storage medium.
  • the computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of the various embodiments of the present disclosure.
  • the aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., and other media that can store program codes.
  • the embodiments of the present disclosure further provide a processor-readable storage medium, wherein the processor-readable storage medium stores a computer program, and the computer program is used to enable a processor to execute the method described in any one of the above method embodiments.
  • the processor-readable storage medium can be any available medium or data storage device that can be accessed by the computer, including but not limited to magnetic storage (such as floppy disks, hard disks, magnetic tapes, magneto-optical disks (MO), etc.), optical storage (such as CD, DVD, BD, HVD, etc.), and semiconductor storage (such as ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)), etc.
  • magnetic storage such as floppy disks, hard disks, magnetic tapes, magneto-optical disks (MO), etc.
  • optical storage such as CD, DVD, BD, HVD, etc.
  • semiconductor storage such as ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)
  • the embodiments of the present disclosure further provide a computer program product, including a computer program, which implements the method described in any one of the above method embodiments when the computer program is executed by a processor.
  • the embodiments of the present disclosure may be provided as methods, systems, or computer programs. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.
  • each process and/or box in the flowchart and/or block diagram, as well as the combination of the process and/or box in the flowchart and/or block diagram can be implemented by computer executable instructions.
  • These computer executable instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for implementing the functions specified in one process or multiple processes in the flowchart and/or one box or multiple boxes in the block diagram.
  • processor-executable instructions may also be stored in a processor-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the processor-readable memory produce a product including an instruction device that implements the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.
  • processor-executable instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本公开提供一种模型同步方法、装置、设备及存储介质,该方法包括:第一设备确定模型调整时刻,根据模型调整时刻,将第一AI模型调整为第二AI模型。多个使用同一AI模型的设备可以同时调整模型,进而可以使多个使用同一AI模型的设备在同一时刻使用的AI模型完全匹配。

Description

一种模型同步方法、装置、设备及存储介质
本公开要求于2022年12月16日提交中国专利局、申请号为202211622245.9、申请名称为“一种模型同步方法、装置、设备及存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本公开中。
技术领域
本公开涉及通信技术领域,具体涉及一种模型同步方法、装置、设备及存储介质。
背景技术
随着人工智能(artificial intelligence,AI)技术和机器学习(Machine Learning,ML)的发展,可以使用AI和/或ML模型(以下简称AI模型)来提升通信系统的性能。
目前,AI模型可能有单边模型或者双边模型,其中,双边模型是指将AI模型分为部分A和部分B,并将部分A和部分B分别部署在两个设备,部分A和部分B需要共同执行才能进行正确的推导过程。然而,目前无法对部分A和部分B进行同步调整操作,可能会导致两个设备在同一时刻使用的模型不匹配,进而导致推导失败。
发明内容
本公开涉及一种模型同步方法、装置、设备及存储介质。
第一方面,本公开实施例提供一种模型同步方法,应用于通信系统中的第一设备,所述通信系统还包括第二设备,所述方法包括:
确定所述第一设备和所述第二设备同步进行模型调整的模型调整时刻;
根据所述模型调整时刻,将第一AI模型调整为第二AI模型。
在一种实施方式中,所述确定模型调整时刻,包括:
获取模型调整信息;
根据所述模型调整信息,确定所述模型调整时刻。
在一种实施方式中,所述模型调整信息包括:模型同步周期和模型调整时长;
所述根据所述模型调整信息,确定所述模型调整时刻,包括:
根据所述模型同步周期和所述模型调整时长,确定所述模型调整时刻。
在一种实施方式中,所述根据所述模型同步周期和所述模型调整时长,确定所述模型调整时刻,包括:
获取指示信息;
根据所述指示信息、所述模型同步周期和所述模型调整时长,确定所述模型调整时刻。
在一种实施方式中,所述模型调整信息包括预设时刻和偏移时长;
所述根据所述模型调整信息,确定模型调整时刻,包括:
根据所述预设时刻和所述偏移时长,确定所述模型调整时刻。
在一种实施方式中,所述模型调整信息还包括:参考时间方式和/或模型同步时间精度。
在一种实施方式中,所述通信系统还包括第三方管理设备。
在一种实施方式中,所述模型调整信息是通过所述第一设备、所述第二设备或者所述第三方管理设备配置;
或者,所述模型调整信息是通过所述第一设备和所述第二设备协商确定;
或者,所述模型调整信息是通过所述第三方管理设备和所述第一设备协商确定;
或者,所述模型调整信息是通过所述第三方管理设备和所述第二设备协商确定;
或者,所述模型调整信息是通过所述第三方管理设备、所述第一设备和所述第二设备协商确定。
在一种实施方式中,所述将第一AI模型调整为第二AI模型,包括以下任意一项:
删除所述第一AI模型,并添加所述第二AI模型;
将所述第一AI模型更换、更新或修改为所述第二AI模型;
去激活所述第一AI模型,并激活所述第二AI模型;
停用所述第一AI模型,并启用所述第二AI模型;
启用所述第二AI模型;
添加所述第二AI模型;
激活所述第二AI模型。
在一种实施方式中,所述方法还包括:
根据初传时刻和所述模型调整时刻,将所述第一AI模型或所述第二AI模型确定为处理第一数据的目标AI模型,所述初传时刻为第二数据的初始传输时刻,所述第二数据为所述目标AI模型处理所述第一数据后得到的数据。
在一种实施方式中,所述根据初传时刻和所述模型调整时刻,将所述第一AI模型或所述第二AI模型确定为处理第一数据的目标AI模型,包括:
若所述初传时刻在所述模型调整时刻之前,则将所述第一AI模型确定为所述目标AI模型;
若所述初传时刻在所述模型调整时刻之后,则将所述第二AI模型确定为所述目标AI模型。
在一种实施方式中,所述第一设备包括终端设备或者网络设备;第二设备包括终端设备或者网络设备。
在一种实施方式中,所述方法还包括:
接收所述第二设备或者所述第三方管理设备发送的所述第二AI模型,所述第二AI模型的接收时刻在所述模型调整时刻之前或者在所述模型调整时刻的确定时刻之前。
在一种实施方式中,所述方法还包括:
向所述第二设备发送所述第二AI模型。
第二方面,本公开实施例提供一种第一设备,应用于通信系统中,所述通信系统 还包括第二设备,所述第一设备包括存储器,收发机,处理器:
所述存储器,用于存储计算机程序;
所述收发机,用于在所述处理器的控制下收发数据;
所述处理器,用于读取所述存储器中的计算机程序并执行如下操作:
确定所述第一设备和第二设备同步进行模型调整的模型调整时刻;
根据所述模型调整时刻,将第一AI模型调整为第二AI模型。
在一种实施方式中,所述处理器,具体用于执行如下操作:
获取模型调整信息;
根据所述模型调整信息,确定所述模型调整时刻。
在一种实施方式中,所述模型调整信息包括:模型同步周期和模型调整时长;
所述处理器,具体用于执行如下操作:
根据所述模型同步周期和所述模型调整时长,确定所述模型调整时刻。
在一种实施方式中,所述处理器,具体用于执行如下操作:
获取指示信息;
根据所述指示信息、所述模型同步周期和所述模型调整时长,确定所述模型调整时刻。
在一种实施方式中,所述模型调整信息包括预设时刻和偏移时长;
所述处理器,具体用于执行如下操作:
根据所述预设时刻和所述偏移时长,确定所述模型调整时刻。
在一种实施方式中,所述模型调整信息还包括:参考时间方式和/或模型同步时间精度。
在一种实施方式中,所述通信系统还包括第三方管理设备。
在一种实施方式中,所述模型调整信息是通过所述第一设备、所述第二设备或者所述第三方管理设备配置;
或者,所述模型调整信息是通过所述第一设备和所述第二设备协商确定;
或者,所述模型调整信息是通过所述第三方管理设备和所述第一设备协商确定;
或者,所述模型调整信息是通过所述第三方管理设备和所述第二设备协商确定;
或者,所述模型调整信息是通过所述第三方管理设备、所述第一设备和所述第二设备协商确定。
在一种实施方式中,所述处理器,具体用于执行如下操作中的任意一项:
删除所述第一AI模型,并添加所述第二AI模型;
将所述第一AI模型更换、更新或修改为所述第二AI模型;
去激活所述第一AI模型,并激活所述第二AI模型;
停用所述第一AI模型,并启用所述第二AI模型;
启用所述第二AI模型;
添加所述第二AI模型;
激活所述第二AI模型。
在一种实施方式中,所述处理器,还用于执行以下操作:
根据初传时刻和所述模型调整时刻,将所述第一AI模型或所述第二AI模型确定为处理第一数据的目标AI模型,所述初传时刻为第二数据的初始传输时刻,所述第二数据为所述目标AI模型处理所述第一数据后得到的数据。
在一种实施方式中,所述处理器,具体用于执行如下操作
若所述初传时刻在所述模型调整时刻之前,则将所述第一AI模型确定为所述目标AI模型;
若所述初传时刻在所述模型调整时刻之后,则将所述第二AI模型确定为所述目标AI模型。
在一种实施方式中,所述处理器,还用于执行以下操作:
接收所述第二设备或者所述第三方管理设备发送的所述第二AI模型,所述第二AI模型的接收时刻在所述模型调整时刻之前或者在所述模型调整时刻的确定时刻之前。
在一种实施方式中,所述处理器,还用于执行以下操作:
向所述第二设备发送所述第二AI模型。
第三方面,本公开实施例提供一种模型同步装置,包括:
第一确定单元,用于确定第一设备和第二设备同步进行模型调整的模型调整时刻;
调整单元,用于根据所述模型调整时刻,将第一AI模型调整为第二AI模型。
在一种实施方式中,所述第一确定单元具体用于:
获取模型调整信息;
根据所述模型调整信息,确定所述模型调整时刻。
在一种实施方式中,所述模型调整信息包括:模型同步周期和模型调整时长;所述第一确定单元具体用于:
根据所述模型同步周期和所述模型调整时长,确定所述模型调整时刻。
在一种实施方式中,所述第一确定单元具体用于:
获取指示信息;
根据所述指示信息、所述模型同步周期和所述模型调整时长,确定所述模型调整时刻。
在一种实施方式中,所述模型调整信息包括预设时刻和偏移时长;所述第一确定单元具体用于:
根据所述预设时刻和所述偏移时长,确定所述模型调整时刻。
在一种实施方式中,所述模型调整信息还包括:参考时间方式和/或模型同步时间精度。
在一种实施方式中,所述通信系统还包括第三方管理设备。
在一种实施方式中,所述模型调整信息是通过所述第一设备、所述第二设备或者所述第三方管理设备配置;
或者,所述模型调整信息是通过所述第一设备和所述第二设备协商确定;
或者,所述模型调整信息是通过所述第三方管理设备和所述第一设备协商确定;
或者,所述模型调整信息是通过所述第三方管理设备和所述第二设备协商确定;
或者,所述模型调整信息是通过所述第三方管理设备、所述第一设备和所述第二设备协商确定。
在一种实施方式中,所述调整单元具体用于实现以下任意一项:
删除所述第一AI模型,并添加所述第二AI模型;
将所述第一AI模型更换、更新或修改为所述第二AI模型;
去激活所述第一AI模型,并激活所述第二AI模型;
停用所述第一AI模型,并启用所述第二AI模型;
启用所述第二AI模型;
添加所述第二AI模型;
激活所述第二AI模型。
在一种实施方式中,所述装置还包括第二确定单元,所述第二确定单元用于:
根据初传时刻和所述模型调整时刻,将所述第一AI模型或所述第二AI模型确定为处理第一数据的目标AI模型,所述初传时刻为第二数据的初始传输时刻,所述第二数据为所述目标AI模型处理所述第一数据后得到的数据。
在一种实施方式中,所述第二确定单元具体用于:
若所述初传时刻在所述模型调整时刻之前,则将所述第一AI模型确定为所述目标AI模型;
若所述初传时刻在所述模型调整时刻之后,则将所述第二AI模型确定为所述目标AI模型。
在一种实施方式中,所述装置还包括接收单元,所述接收单元用于:
接收所述第二设备或者所述第三方管理设备发送的所述第二AI模型,所述第二AI模型的接收时刻在所述模型调整时刻之前或者在所述模型调整时刻的确定时刻之前。
在一种实施方式中,所述装置还包括发送单元,所述发送单元用于:
向所述第二设备发送所述第二AI模型。
第四方面,本公开实施例提供一种处理器可读存储介质,所述处理器可读存储介质存储有计算机程序,所述计算机程序用于使所述处理器执行第一方面所述的方法。
本公开提供一种模型同步方法、装置、设备及存储介质,该方法中所有使用同一AI模型的设备先确定模型调整时刻,每个使用同一AI模型的设备再根据模型调整时刻,同时将第一AI模型调整为第二AI模型。多个使用同一AI模型的设备可以同时调整模型,进而可以使多个使用同一AI模型的设备在同一时刻使用的AI模型完全匹配。
应当理解,上述发明内容部分中所描述的内容并非旨在限定本公开的实施例的关键或重要特征,亦非用于限制本公开的范围。本公开的其它特征将通过以下的描述变得容易理解。
附图说明
为了更清楚地说明本公开或现有技术中的技术方案,下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图是本发明的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动性的前提下,还可以根据这些附图获得其他的附图。
图1为本公开实施例提供的通信系统的一种架构示意图;
图2为本公开实施例提供的一种模型同步方法的流程图;
图3为本公开实施例提供的协商模型调整信息的流程图一;
图4为本公开实施例提供的协商模型调整信息的流程图二;
图5为本公开实施例提供的一种确定模型调整时刻的示意图;
图6为本公开实施例提供的另一种模型同步方法的示意图;
图7为本公开实施例提供的确定目标AI模型的示意图;
图8A为本公开实施例提供的第一设备的结构示意图一;
图8B为本公开实施例提供的第一设备的结构示意图二;
图9A为本公开实施例提供的一种模型同步装置的结构示意图一;
图9B为本公开实施例提供的一种模型同步装置的结构示意图二;
图9C为本公开实施例提供的一种模型同步装置的结构示意图三;
图9D为本公开实施例提供的一种模型同步装置的结构示意图四。
具体实施方式
本公开实施例中术语“和/或”,描述关联对象的关联关系,表示可以存在三种关系,例如,A和/或B,可以表示:单独存在A,同时存在A和B,单独存在B这三种情况。字符“/”一般表示前后关联对象是一种“或”的关系。
本公开实施例中术语“多个”是指两个或两个以上,其它量词与之类似。
下面将结合本公开实施例中的附图,对本公开实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本公开一部分实施例,并不是全部的实施例。基于本公开中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都属于本公开保护的范围。
本公开实施例提供一种模型同步方法,多个使用AI模型的设备可以同时调整模型,进而可以使多个使用AI模型的设备在同一时刻使用的AI模型完全匹配。
其中,方法和装置是基于同一申请构思的,由于方法和装置解决问题的原理相似,因此装置和方法的实施可以相互参见,重复之处不再赘述。
本公开实施例提供的技术方案可以适用于多种系统,尤其是5G系统。例如适用的系统可以是全球移动通讯(global system of mobile communication,GSM)系统、码分多址(code division multiple access,CDMA)系统、宽带码分多址(wideband code division multiple access,WCDMA)通用分组无线业务(general packet radio service,GPRS)系统、长期演进(long term evolution,LTE)系统、LTE频分双工(frequency division duplex,FDD)系统、LTE时分双工 (time division duplex,TDD)系统、高级长期演进(long term evolution advanced,LTE-A)系统、通用移动系统(universal mobile telecommunication system,UMTS)、全球互联微波接入(worldwide interoperability for microwave access,WiMAX)系统、5G新空口(new radio,NR)系统等。这多种系统中均包括终端设备和网络设备。系统中还可以包括核心网部分,例如演进的分组系统(evloved packet system,EPS)、5G系统(5GS)等。
本公开实施例涉及的终端设备,可以是指向用户提供语音和/或数据连通性的设备,具有无线连接功能的手持式设备、或连接到无线调制解调器的其他处理设备等。在不同的系统中,终端设备的名称可能也不相同,例如在5G系统中,终端设备可以称为用户设备(user equipment,UE)。无线终端设备可以经无线接入网(radio access network,RAN)与一个或多个核心网(core network,CN)进行通信,无线终端设备可以是移动终端设备,如移动电话(或称为“蜂窝”电话)和具有移动终端设备的计算机,例如,可以是便携式、袖珍式、手持式、计算机内置的或者车载的移动装置,它们与无线接入网交换语言和/或数据。例如,个人通信业务(personal communication service,PCS)电话、无绳电话、会话发起协议(session initiated protocol,SIP)话机、无线本地环路(wireless local loop,WLL)站、个人数字助理(personal digital assistant,PDA)等设备。无线终端设备也可以称为系统、订户单元(subscriber unit)、订户站(subscriber station),移动站(mobile station)、移动台(mobile)、远程站(remote station)、接入点(access point)、远程终端设备(remote terminal)、接入终端设备(access terminal)、用户终端设备(user terminal)、用户代理(user agent)、用户装置(user device),本公开实施例中并不限定。
本公开实施例涉及的网络设备,可以是接入网设备,也可以是核心网设备。
其中,接入网设备可以是基站,该基站可以包括多个为终端提供服务的小区。根据具体应用场合不同,基站又可以称为接入点,或者可以是接入网中在空中接口上通过一个或多个扇区与无线终端设备通信的设备,或者其它名称。网络设备可用于将收到的空中帧与网际协议(internet protocol,IP)分组进行相互更换,作为无线终端设备与接入网的其余部分之间的路由器,其中接入网的其余部分可包括网际协议(IP)通信网络。网络设备还可协调对空中接口的属性管理。例如,本公开实施例涉及的网络设备可以是全球移动通信系统(global system for mobile communications,GSM)或码分多址接入(code division multiple access,CDMA)中的网络设备(base transceiver station,BTS),也可以是带宽码分多址接入(wide-band code division multiple access,WCDMA)中的网络设备(NodeB),还可以是长期演进(long term evolution,LTE)系统中的演进型网络设备(evolutional Node B,eNB或e-NodeB)、5G网络架构(next generation system)中的5G基站(gNB),也可以是家庭演进基站(Home evolved Node B,HeNB)、中继节点(relay node)、家庭基站(femto)、微微基站(pico)等,本公开实施例中并不限定。在一些网络结构中,网络设备可以包括集中单元(centralized unit,CU)节点和分布单元(distributed unit,DU)节点,集中单元和分布单元也可以地理上分开布置。
核心网设备可以包括如下任意一项:网络数据分析功能(network data analytics function,NWDAF)实体,策略控制功能(policy control function,PCF)实体,应用功能(application  function,AF)实体,接入和移动性管理功能(access and mobility management function,AMF)实体,会话管理功能(session management function,SMF)实体,网络开放功能(network exposure function,NEF)实体,用户面功能(user plane function,UPF)实体,统一数据库(unified data repository,UDR)实体,网络切片选择功能(network slice selection function,NSSF)实体,认证服务器功能(authentication server function,AUSF)实体,统一数据管理(unified data management,UDM)实体,网络功能数据库功能(network repository function,NRF)实体。
下面结合图1对本公开的通信场景进行说明。图1为本公开实施例提供的通信系统的一种架构示意图。
如图1所示,该架构包括终端设备、网络设备和第三方管理设备。其中,终端设备、网络设备和第三方管理设备均可以作为AI模型改变的发起方,终端设备和网络设备均可以作为AI模型使用方。
在当前标准协议中对于双边(two-sided)AI模型,一个AI模型分布在多个不同的设备,共同完成AI模型推导(model inference),而由于需要在不同设备共同执行推导,需要多个设备的模型部分完全同步匹配,归属于同一个AI模型。当前传统的无线资源控制(radio resource control,RRC)信令传递过程,需要设备A给另一个设备B发送模型,或指示应用哪个模型,设备B接收到信令后进行解析并启用新模型,对于双边AI模型无法做到两侧在相同的时刻启用模型,可能造成因为模型改变/转换过程中的时间差导致的模型使用不匹配问题,最终导致AI推导失败。
基于现有技术中的问题,本公开提出了如下技术构思:所有使用同一AI模型的设备先确定模型调整时刻,每个使用同一AI模型的设备再根据模型调整时刻,同时将第一AI模型调整为第二AI模型。多个使用同一AI模型的设备可以同时调整模型,进而可以使多个使用同一AI模型的设备在同一时刻使用的AI模型完全匹配。
下面结合具体的实施例对本公开提供的AI模型同步方法进行介绍。
图2为本公开实施例提供的一种模型同步方法的流程图。如图2所示,该方法包括:
S201、确定第一设备和第二设备同步进行模型调整的模型调整时刻。
第一设备和第二设备是通信系统中的设备。
第一设备可以是终端设备,也可以是网络设备。
第二设备可以是终端设备,也可以是网络设备。
第一设备和第二设备均可以作为模型改变的发起方,也可以作为模型使用方。
在通信系统中还可以包括第三方管理设备,第三方管理设备可以作为模型改变的发起方。
在一种可能的实现中,可以通过以下方式确定模型调整时刻:获取模型调整信息,根据模型调整信息,确定模型调整时刻。
在一种可能的实现中,模型调整信息包括以下至少一项:模型同步周期、模型调整时长、预设时刻、偏移时长、参考时间方式、模型同步时间精度。
在一种可能的实现中,模型调整信息可以包括以下任意一项:
模型同步周期和模型调整时长;
模型同步周期、模型调整时长和预设时刻;
预设时刻和偏移时长;
模型同步周期、模型调整时长和参考时间方式;
模型同步周期、模型调整时长和模型同步时间精度;
模型同步周期、模型调整时长、参考时间方式和模型同步时间精度;
预设时刻、偏移时长和参考时间方式;
预设时刻、偏移时长和模型同步时间精度;
预设时刻、偏移时长、参考时间方式和模型同步时间精度。
模型同步周期可以通过模型改变的发起方预先设置,也可以通过模型改变的发起方监听系统的性能,根据监听到的系统的性能结果确定模型同步周期。例如,若监听到的系统的性能结果较好,则可以设置一个较长的模型同步周期;若监听到的系统的性能结果较差时,则可以设置一个较短的模型同步周期。
模型调整时长可以是指模型调整时刻与模型同步周期的起始时刻之间的偏移时长。在一种可能的实现中,模型调整时长小于模型同步周期的时长。
预设时刻可以是指绝对时刻,预设时刻可以是指模型调整时刻,也可以是指任意一个绝对时刻。
偏移时长可以是指模型调整时刻与预设时刻之间的偏移时长。
参考时间方式是指多个模型使用方进行初始时间同步的参考时间方式,例如,可以通过全球定位系统(global positioning system,GPS)+协调世界时(universal time coordinated,UTC)方式,或者1588方式达到多个模型使用方的初始时间同步。
模型同步时间精度,可以用于决定模型调整时刻的精度。示例性的,模型同步时间精度可以是时隙(slot)级别的模型同步时间精度、子帧(subframe)级别的模型同步时间精度或者帧(frame)级别的模型同步时间精度;针对不同用例的模型部署,或者针对不同AI模型改变操作,可以要求不同的模型同步时间精度,例如针对信道状态信息(channel state information,CSI)反馈可以要求时隙级别的模型同步时间精度,而定位可以要求帧级别的模型同步时间精度。又例如对于更换/激活/去激活/更新,可以要求较高的模型同步时间精度(例如时隙级别),从而使得使用的模型完全同步;对于模型的添加/删除/修改,可能要求较低的模型同步时间精度(例如帧级别),因为模型添加之后还需要更高模型同步时间精度的激活功能,才能真正使用该AI模型。
在一种可能的实现中,可以通过以下方式获取模型调整信息:
模型改变的发起方配置模型调整信息,并向模型使用方指示模型调整信息;或者,模型改变的发起方与模型使用方协商确定模型调整信息。
S202a、第一设备根据模型调整时刻,将第一AI模型调整为第二AI模型。
在一种可能的实现中,AI模型可以包括AI模型和/或ML模型。
在一种可能的实现中,可以通过以下任意一种方式将第一AI模型调整为第二AI模型:
删除第一AI模型,并添加第二AI模型;
将第一AI模型更换、更新或修改为第二AI模型;
去激活第一AI模型,并激活第二AI模型;
停用第一AI模型,并启用第二AI模型;
启用第二AI模型;
添加第二AI模型;
激活第二AI模型;
在一种可能的实现中,第二AI模型可以由模型改变的发起方确定,并由模型改变的发起方向模型使用方发送第二AI模型。
在一种可能的实现中,可以通过以下任意一种方式确定和发送第二AI模型:
1、第三方管理设备确定第二AI模型,并向第一设备发送第二AI模型;
2、第三方管理设备确定第二AI模型,并向第二设备发送第二AI模型;
3、第一设备确定第二AI模型,并向第二设备发送第二AI模型;
4、第二设备确定第二AI模型,并向第一设备发送第二AI模型。
针对第1种方式,可以细化为以下任意一种方式:
1.1、第三方管理设备确定第二AI模型,并向多个终端设备发送第二AI模型;
1.2、第三方管理设备确定第二AI模型,并向多个网络设备发送第二AI模型;
1.3、第三方管理设备确定第二AI模型,并向多个终端设备和网络设备发送第二AI模型。
第2种方式的细化方式可以参见第1种方式的细化方式,此处不再赘述。
针对第3种方式,可以细化为以下任意一种方式:
3.1、第一终端设备确定第二AI模型,并向多个第二终端设备发送第二AI模型;
3.2、终端设备确定第二AI模型,并向多个网络设备发送第二AI模型;
3.3、第一终端设备确定第二AI模型,并向多个第二终端设备和网络设备发送第二AI模型;
3.4、第一网络设备确定第二AI模型,并向多个第二网络设备发送第二AI模型;
3.5、网络设备确定第二AI模型,并向多个终端设备发送第二AI模型;
3.6、第一网络设备确定第二AI模型,并向多个第二网络设备和终端设备发送第二AI模型。
第4种方式的细化方式可以参见第3种方式的细化方式,此处不再赘述。
在一种可能的实现中,模型改变的发起方需要在模型调整时刻之前确定第二AI模型;模型使用方需要在模型调整时刻之前或者在模型调整时刻的确定时刻之前接收第二AI模型。
在一种可能的实现中,将第一AI模型更换为第二AI模型可以是指删除第一AI模型,并添加第二AI模型。
在一种可能的实现中,将第一AI模型更新为第二AI模型可以是指删除第一AI模型,并添加第二AI模型;也可以是在第一AI模型的基础上,对第一AI模型的代码进行添加、删除或修改后,得到第二AI模型。
在一种可能的实现中,去激活第一AI模型可以是指终止第一AI模型的激活状态。
在一种可能的实现中,可以通过以下方式实现停用第一AI模型:
模型使用方停用或启用整体AI功能/用例,由于整个AI功能/用例中包括一个或者多个AI模型,因此,若整个AI功能/用例停用,则整个AI功能/用例中的所有AI模型均被停用,若整个AI功能/用例启用,则整个AI功能/用例中的所有AI模型均被启用。
在一种可能的实现中,停用第一AI模型后,若不启用第二AI模型,可以回退至传统功能,例如,CSI反馈(feedback)回退至基本的码本参数上报。
在一种可能的实现中,模型使用方中的每个AI用例/功能可以包括一个或者多个AI模型,同一个AI用例/功能之间的多个AI模型之间可以进行切换,多个或者单个模型可以进行添加、删除、修改、更换、更新、激活和/或去激活操作。
在一种可能的实现中,第一设备可以在模型调整时刻将第一AI模型调整为第二AI模型。
在一种可能的实现中,也可以通过不同模型使用方传递的数据中间结果的格式/形式判断什么时候进行模型调整。也就是说数据接收方可以将采用AI模型推导得到的数据中间结果的数据格式/形式与采用传统码本方式得到的数据中间结果的格式/形式比较,再根据两者的数据中间结果的格式/形式判断在接收方是否执行模型调整。
S202b、第二设备根据模型调整时刻,将第一AI模型调整为第二AI模型。
在一种可能的实现中,第二设备中的第一AI模型和第一设备中的第一AI模型属于同一AI模型;第二设备中的第二AI模型和第一设备中的第二AI模型也属于同一AI模型;只是各个设备中包括的是同一模型的不同部分。
示例性的、CSI反馈的AI增强功能中的空频域CSI压缩子用例(spatial-frequency domain CSI compression),完整的AI模型由部分A与部分B共同组成。AI模型部分A用于在终端设备侧进行基于AI模型的编码推导,AI模型部分B用于在网络设备侧进行基于AI模型的解码推导。
第一设备和第二设备均可以根据同一模型调整时刻对模型进行调整。即各个模型使用方需要同时对同一AI模型的不同部分进行模型调整。
本公开实施例提供的模型同步方法,包括:第一设备确定第一设备和第二设备同步进行模型调整的模型调整时刻;并根据模型调整时刻,将第一AI模型调整为第二AI模型。同时第二设备可以根据模型调整时刻,将第一AI模型调整为第二AI模型。即所有使用同一AI模型的设备先确定模型调整时刻,每个使用同一AI模型的设备再根据模型调整时刻同时进行模型调整,进而可以使多个使用同一AI模型的设备在同一时刻使用的AI模型完全匹配。
在图2所示实施例的基础上,下面,详细说明模型使用方如何获取模型调整信息。
在一种可能的实现中,可以通过以下任意一种方式获取模型调整信息:
1、第三方管理设备配置模型调整信息,多个模型使用方接收第三方管理设备发送的模型调整信息;
2、第一设备配置模型调整信息,多个模型使用方接收第一设备发送的模型调整信息;
3、第二设备配置模型调整信息,多个模型使用方接收第二设备发送的模型调整信息;
4、第一设备和第二设备协商模型调整信息;
5、第三方管理设备和第一设备协商模型调整信息;
6、第三方管理设备和第二设备协商模型调整信息;
7、第三方管理设备、第一设备和第二设备协商模型调整信息。
模型使用方可以是终端设备也可以是网络设备。
当模型使用方接收到模型改变的发起方发送的模型调整信息后,还可以向模型改变的发起方发送确认信息。
针对第1种方式,可以细化为以下任意一种方式:
1.1、第三方管理设备配置模型调整信息,多个终端设备接收第三方管理设备发送的模型调整信息;
1.2、第三方管理设备配置模型调整信息,多个网络设备接收第三方管理设备发送的模型调整信息;
1.3、第三方管理设备配置模型调整信息,多个终端设备和网络设备接收第三方管理设备发送的模型调整信息。
针对第2种方式,可以细化为以下任意一种方式:
2.1、第一网络设备配置模型调整信息,多个第二网络设备接收第一网络设备发送的模型调整信息;
2.2、网络设备配置模型调整信息,多个终端设备接收第一网络设备发送的模型调整信息;
2.3、第一网络设备配置模型调整信息,多个终端设备和第二网络设备接收第一网络设备发送的模型调整信息。
2.4、终端设备配置模型调整信息,多个网络设备接收终端设备发送的模型调整信息;
2.5、第一终端设备配置模型调整信息,多个第二终端设备接收第一终端设备发送的模型调整信息;
2.6、第一终端设备配置模型调整信息,多个第二终端设备和网络设备接收第一终端设备发送的模型调整信息。
第3种方式的细化方式可以参见第2种方式的细化方式,此处不再赘述。
针对第4种方式,可以细化为以下任意一种方式:
4.1、第一终端设备和第二终端设备协商模型调整信息。
4.2、网络设备和终端设备协商模型调整信息;
4.3、第一网络设备和第二网络设备协商模型调整信息。
针对第5种方式,可以细化为以下任意一种方式:
5.1、第三方管理设备和终端设备协商模型调整信息;
5.2、第三方管理设备和网络设备协商模型调整信息。
第6种方式的细化方式可以参见第5种方式的细化方式,此处不再赘述。
针对第7种方式,可以细化为以下任意一种方式:
7.1、第三方管理设备、第一终端设备和第二终端设备协商模型调整信息;
7.2、第三方管理设备、第一网络设备和第二网络设备协商模型调整信息;
7.3、第三方管理设备、终端设备和网络设备协商模型调整信息。
下面,结合图3,详细说明设备间如何协商模型调整信息。
图3为本公开实施例提供的协商模型调整信息的流程图一。如图3所示,以模型改变的发起方为网络设备,模型使用方为网络设备和终端设备为例,该方法包括:
S301、网络设备确定第一模型调整信息。
第一模型调整信息与图2所示实施例中的模型调整信息的内容相同,只是具体的参数值和/或类型不一样。
为了减少协商次数,缩短协商时长,网络设备可以根据系统的性能确定第一模型调整信息。
S302、网络设备向终端设备发送第一模型调整信息。
网络设备向终端设备发送第一模型调整信息,也可以表示为终端设备接收网络设备发送的第一模型调整信息。
终端设备接收到第一模型调整信息后,可以根据第一模型调整信息确定第一模型调整时刻。若终端设备确定自身无法在第一模型调整时刻进行模型调整,则可以执行S303。
S303、终端设备确定第二模型调整信息。
第二模型调整信息与第一模型调整信息的内容相同,只是具体的参数值和/或类型不一样。
根据第二模型调整信息确定的第二模型调整时刻与根据第一模型调整信息确定的第一模型调整时刻不同。
S304、终端设备向网络设备发送第二模型调整信息。
终端设备向网络设备发送第二模型调整信息,也可以表示为网络设备接收终端设备发送的第二模型调整信息。
网络设备接收到第二模型调整信息后,可以根据第二模型调整信息确定第二模型调整时刻,并在第二模型调整时刻与终端设备一起进行模型调整。
需要说明的是,网络设备与终端设备之间的协商次数可以不局限于图3所示实施例示出的次数。
例如,网络设备在接收第二模型调整信息后,若确定自身无法在第二模型调整时刻进行模型调整,可以重新确定第三模型调整信息,并发送给终端设备,以此类推,直至网络设备和终端设备协商出双方均认可的模型调整信息。但是为了减少协商时长,应尽量减少协商次数。
上述4、5和6列举的设备之间协商模型调整信息的方式可以参见图3所示实施例中示出的协商方式。
下面,结合图4,详细说明第三方管理设备、第一设备和第二设备之间如何协商模型调整信息。
图4为本公开实施例提供的协商模型调整信息的流程图二。如图4所示,所述方法包 括:
S401、第三方管理设备确定第一模型调整信息。
第一模型调整信息与图2所示实施例中的模型调整信息的内容相同,只是具体的参数值和/或类型不一样。
为了减少协商次数,缩短协商时长,第三方管理设备可以根据系统的性能确定第一模型调整信息。
S402、第三方管理设备分别向第一设备和第二设备发送第一模型调整信息。
第三方管理设备分别向第一设备和第二设备发送第一模型调整信息,也可以表示为第一设备接收第三方管理设备发送的第一模型调整信息;第二设备接收第三方管理设备发送的第一模型调整信息。
第一设备和第二设备接收到第一模型调整信息后,可以根据第一模型调整信息确定第一模型调整时刻。若第一设备(或者第二设备)确定自身无法在第一模型调整时刻进行模型调整,则可以执行S403。
S403、第一设备(或者第二设备)确定第二模型调整信息。
第二模型调整信息与第一模型调整信息的内容相同,只是具体的参数值和/或类型不一样。
根据第二模型调整信息确定的第二模型调整时刻与根据第一模型调整信息确定的第一模型调整时刻不同。
S404、第一设备(或者第二设备)向第三方管理设备发送第二模型调整信息。
第一设备(或者第二设备)向第三方管理设备发送第二模型调整信息,也可以表示为第三方管理设备接收第一设备(或者第二设备)发送的第二模型调整信息。
S405、第三方管理设备向第二设备(或第一设备)发送第二模型调整信息。
第三方管理设备向第二设备(或第一设备)发送第二模型调整信息,也可以表示为第二设备(或第一设备)接收第三方管理设备发送的第二模型调整信息。
第二设备(或第一设备)接收到第二模型调整信息后,可以根据第二模型调整信息确定第二模型调整时刻,并在第二模型调整时刻与第一设备(或第二设备)一起进行模型调整。
在一种可能的实现中,在进行AI模型同步之前,使用同一个AI模型的多个设备可以先进行网络同步,后续才能够正常的进行数据传输。即在网络同步的基础上,后续才能够进行有效的AI模型同步。
网络同步的方式可以参见现有技术中的方式,此处不再赘述。
在上述任意实施例的基础上,下面结合图5和图6详细说明如何根据模型调整信息确定模型调整时刻。
在一种可能的实现中,可以根据模型同步周期和模型调整时长,确定模型调整时刻。
在一种可能的实现中,可以确定模型同步周期的起始时刻,根据模型同步周期的起始时刻和模型调整时长,确定模型调整时刻。若模型同步周期的起始时刻为ts,模型调整时长为TL,则模型调整时刻为ts+TL。
在一种可能的实现中,模型同步周期的起始时刻可以根据第二AI模型的接收时刻确定模型同步周期的起始时刻;也可以根据模型调整信息的接收时刻确定模型同步周期的起始时刻;还可以由模型改变的发起方指示模型同步周期的起始时刻。
在一种可能的实现中,可以通过以下方式确定模型调整时刻:
模型改变的发起方向模型使用方发送指示信息,模型使用方根据指示信息、模型同步周期和模型调整时长,确定模型调整时刻。
在一种可能的实现中,模型使用方可以根据指示信息和模型同步周期确定目标模型同步周期,根据目标模型同步周期的起始时刻和模型调整时长,确定模型调整时刻。
目标模型同步周期可以是指进行模型调整的周期。
例如,模型使用方根据模型同步周期可以确定N(N>0)个模型同步周期,根据N个模型同步周期和模型调整时长,可以确定N个待选模型调整时刻。若是在第i个模型同步周期接收到指示信息,若指示信息的接收时刻在第i个待选模型调整时刻之前,则可以将第i个模型同步周期确定为目标模型同步周期,将第i个待选模型调整时刻确定为目标模型调整时刻,并在目标模型调整时刻进行模型调整;若指示信息的接收时刻在第i个待选模型调整时刻之后,则可以将第i+1个模型同步周期确定为目标模型同步周期,将第i个待选模型调整时刻确定为目标模型调整时刻,并在目标模型调整时刻进行模型调整,其中,i=1、2、…、N。
为了便于理解,下面,结合图5,详细说明如何确定模型调整时刻。
图5为本公开实施例提供的一种确定模型调整时刻的示意图。如图5所示,以网络设备为模型改变的发起方,终端设备为模型使用方为例。终端设备和网络设备在t0时刻进行网络同步;然后终端设备在t1时刻接收网络设备发送的模型调整信息,模型调整信息中包括模型调整周期、模型调整时长和模型同步时间精度,其中,模型调整周期为200ms、模型调整时长为5个时隙,模型同步时间精度为时隙级别;终端设备根据模型调整信息确定多个模型改变周期。若终端设备在t2时刻接收到网络设备发送的第二AI模型或者指示信息,t2时刻位于第N个模型调整周期,则可以在第N+1个模型调整周期将第一AI模型调整为第二AI模型,具体的模型调整时刻可以根据第N+1个模型调整周期的起始时刻和模型调整时长确认,其中,N>0。
在一种可能的实现中,模型调整时长可以为0,也可以大于0。若模型调整时长为0,则表示在模型同步周期的边界进行模型调整;若模型调整时长大于0,则表示在模型调整周期中进行模型调整。
为了便于理解,下面提供一个具体的示例详细说明模型同步过程。
图6为本公开实施例提供的另一种模型同步方法的示意图。如图6所示,以网络设备作为模型改变的发起方,以终端设备和网络设备作为模型使用房为例,该方法包括:
S601、网络设备和终端设备进行网络同步。
S602、网络设备和终端设备协商模型调整信息。
模型调整信息中包括模型同步周期和模型调整时长。
协商模型调整信息的方式可以参见上述实施例,此处不再赘述。
S603、网络设备向终端设备发送第二AI模型。
S604、网络设备向终端设备发送指示信息。
指示信息用于指示可以开始模型调整。
S605、网络设备和终端设备根据模型调整信息和指示信息,确定模型调整时刻,并在模型调整时刻进行模型调整。
确定模型调整时刻的方式可以参见上述实施例,此处不再赘述。
在一种可能的实现中,模型使用方可以根据预设时刻和偏移时长,确定模型调整时刻。
在一种可能的实现中,偏移时长可以0,也可以大于0。
当偏移时长为0时,则表示可以直接将预设时刻作为模型调整时刻;当偏移时长大于0时,则表示将距离预设时刻之后偏移时长的时刻作为模型调整时刻。
例如,若预设时刻为t,偏移时长为T,则模型调整时刻为t+T。
在上述任意实施例的基础上,下面详细说明各个模型使用方如何确定所使用的目标AI模型。
在一种可能的实现中,模型使用方可以根据初传时刻和模型调整时刻,确定处理第一数据的目标AI模型,初传时刻为第二数据的初始传输时刻,第二数据为目标AI模型处理第一数据后得到的数据。
在一种可能的实现中,模型使用方可以根据初传时刻和模型调整时刻,将第一AI模型或第二AI模型确定为处理第一数据的目标AI模型:若初传时刻在模型调整时刻之前,则将第一AI模型确定为目标AI模型;若初传时刻在模型调整时刻之后,则将第二AI模型确定为目标AI模型。
在一种可能的实现中,若初传时刻在模型调整时刻之前,则无论第二数据是初传还是重传,模型使用方均使用第一AI模型作为目标AI模型;若初传时刻在模型调整时刻之后,则无论第二数据是初传还是重传,模型使用方均使用第二AI模型作为目标AI模型。
若使用物理上行链路控制信道(physical uplink control channel,PUCCH)将第二数据从终端设备传输给网络设备,则没有重传,只有初传。可以认为是在一个时隙内完成终端设备发送与网络设备接收。
若使用物理上行共享信道(physical uplink shared channel,PUSCH)将第二数据从终端设备传输给网络设备,则有可能发生重传。此时,需要先确定初传时刻,再根据初传时刻和模型调整时刻,确定处理第一数据的目标AI模型。
例如,对于CSI反馈用例中,终端设备执行基于AI模型推导的编码操作,并将推导出的数据中间结果(第二数据)发送到网络设备,网络设备执行基于AI模型推导的解码操作:
在接收方接收到第二数据时,需推导发送方在编码时使用的AI模型,首先需要确定第二数据相关信息在终端设备侧进行初始传输的时刻(由于重传与初传均为网络设备侧调度,网络设备侧可获知初传时刻),以确定终端设备侧编码所使用的AI模型;即如果推导出的初传时刻在模型调整时刻之前,则即使已经到达模型调整时刻从而将旧AI模型(第一AI模型)调整为新AI模型(第二AI模型),也需使用旧AI模型进行解码;因此需要 网络设备侧在模型调整后,暂时保存所使用的前一个AI模型(旧AI模型)的信息。如图7所示,终端设备使用旧AI模型对第一数据执行推导,得到第二数据,并在t1时刻初次向网络设备传输第二数据,网络设备在t1时刻对初传的第二数据进行解码,并且解码失败;t2时刻为模型调整时刻,网络设备和终端设备均在t2时刻将旧AI模型调整为新AI模型,并且,网络设备还需要保存旧AI模型;终端设备在t3时刻重复向网络设备传输第二数据,网络设备在t3时刻使用旧AI模型对重传的第二数据执行推导。
在发送方使用AI模型推导要发送的第二数据时,如果确定要在模型调整时刻之后发起初传,则即使编码发生在模型调整时刻之前,也需按照后一个新AI模型进行编码的AI模型推导操作。
准确确定目标AI模型,可以保证各模型使用方使用的模型完全匹配。
图8A为本公开实施例提供的第一设备的结构示意图一。如图8A所示,该第一设备包括:存储器801,收发机802和处理器803。
存储器801,用于存储计算机程序;
收发机802,用于在处理器803的控制下收发数据;
处理器803,用于读取存储器801中存储的计算机程序并执行以下操作:
确定第一设备和第二设备同步进行模型调整的模型调整时刻;
根据模型调整时刻,将第一AI模型调整为第二AI模型。
在一种实施方式中,处理器803,具体用于执行如下操作:
获取模型调整信息;
根据模型调整信息,确定所述模型调整时刻。
在一种实施方式中,模型调整信息包括:模型同步周期和模型调整时长;
处理器803,具体用于执行如下操作:
根据模型同步周期和模型调整时长,确定模型调整时刻。
在一种实施方式中,处理器803,具体用于执行如下操作:
获取指示信息;
根据指示信息、模型同步周期和模型调整时长,确定模型调整时刻。
在一种实施方式中,所述模型调整信息包括预设时刻和偏移时长;
处理器803,具体用于执行如下操作:
根据预设时刻和偏移时长,确定模型调整时刻。
在一种实施方式中,模型调整信息还包括:参考时间方式和/或模型同步时间精度。
在一种实施方式中,通信系统还包括第三方管理设备。
在一种实施方式中,模型调整信息是通过第一设备、第二设备或者第三方管理设备配置;
或者,模型调整信息是通过第一设备和第二设备协商确定;
或者,模型调整信息是通过第三方管理设备和第一设备协商确定;
或者,模型调整信息是通过第三方管理设备和第二设备协商确定;
或者,模型调整信息是通过第三方管理设备、第一设备和第二设备协商确定。
在一种实施方式中,处理器803,具体用于执行如下操作中的任意一项:
删除第一AI模型,并添加第二AI模型;
将第一AI模型更换、更新或修改为第二AI模型;
去激活第一AI模型,并激活第二AI模型;
停用第一AI模型,并启用第二AI模型;
启用第二AI模型;
添加第二AI模型;
激活第二AI模型。
在一种实施方式中,处理器803,还用于执行以下操作:
根据初传时刻和所述模型调整时刻,确定处理第一数据的目标AI模型,所述初传时刻为第二数据的初始传输时刻,所述第二数据为所述目标AI模型处理所述第一数据后得到的数据。
在一种实施方式中,处理器,具体用于执行如下操作:
若初传时刻在模型调整时刻之前,则将第一AI模型确定为目标AI模型;
若初传时刻在模型调整时刻之后,则将第二AI模型确定为目标AI模型。
在一种实施方式中,处理器803,还用于执行以下操作:
接收第二设备或者第三方管理设备发送的第二AI模型,第二AI模型的接收时刻在模型调整时刻之前或者在模型调整时刻的确定时刻之前。
在一种实施方式中,处理器803,还用于执行以下操作:
向第二设备发送第二AI模型。
其中,在图8A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803代表的一个或多个处理器和存储器801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802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,这些传输介质包括无线信道、有线信道、光缆等传输介质。处理器803负责管理总线架构和通常的处理,存储器801可以存储处理器803在执行操作时所使用的数据。
图8B为本公开实施例提供的第一设备的结构示意图二。如图8B所示,当第一设备为终端设备的时候,该设备还可以包括用户接口804,针对不同的终端设备,用户接口804还可以是能够外接内接需要设备的接口,连接的设备包括但不限于小键盘、显示器、扬声器、麦克风、操纵杆等。
可选地,处理器803可以是中央处理器(central processing unit,CPU)、专用集成电路(application specific integrated circuit,ASIC)、现场可编程门阵列(field-programmable gate array,FPGA)或复杂可编程逻辑器件(complex programmable logic device,CPLD),处理器也可以采用多核架构。
处理器803通过调用存储器801存储的计算机程序,用于按照获得的可执行指令执行本公开实施例提供的任一所述方法。处理器803与存储器801也可以物理上分开布置。
在此需要说明的是,本公开提供的上述第一设备,能够实现上述方法实施例中第一设备所实现的所有方法步骤,且能够达到相同的技术效果,在此不再对本实施例中与方法实施例相同的部分及有益效果进行具体赘述。
图9A为本公开实施例提供的一种模型同步装置的结构示意图一。如图9所示,该装置包括:
第一确定单元901,用于确定第一设备和第二设备同步进行模型调整的模型调整时刻;
调整单元902,用于根据模型调整时刻,将第一AI模型调整为第二AI模型。
在一种实施方式中,第一确定单元901,具体用于:
获取模型调整信息;
根据模型调整信息,确定模型调整时刻。
在一种实施方式中,模型调整信息包括:模型同步周期和模型调整时长;第一确定单元901,具体用于:
根据模型同步周期和所述模型调整时长,确定模型调整时刻。
在一种实施方式中,第一确定单元901,具体用于:
获取指示信息;
根据指示信息、模型同步周期和模型调整时长,确定模型调整时刻。
在一种实施方式中,模型调整信息包括预设时刻和偏移时长;第一确定单元901,具体用于:
根据预设时刻和偏移时长,确定模型调整时刻。
在一种实施方式中,模型调整信息还包括:参考时间方式和/或模型同步时间精度。
在一种实施方式中,通信系统还包括第三方管理设备。
在一种实施方式中,模型调整信息是通过第一设备、第二设备或者第三方管理设备配置;
或者,模型调整信息是通过第一设备和第二设备协商确定;
或者,模型调整信息是通过第三方管理设备和第一设备协商确定;
或者,模型调整信息是通过第三方管理设备和第二设备协商确定;
或者,模型调整信息是通过第三方管理设备、第一设备和第二设备协商确定。
在一种实施方式中,调整单元902,具体用于实现以下任意一项:
删除第一AI模型,并添加第二AI模型;
将第一AI模型更换、更新或修改为第二AI模型;
去激活第一AI模型,并激活第二AI模型;
停用第一AI模型,并启用第二AI模型;
启用第二AI模型;
添加第二AI模型;
激活第二AI模型。
图9B为本公开实施例提供的一种模型同步装置的结构示意图二。如图9B所示,装置还包括第二确定单元903,第二确定单元903用于:
根据初传时刻和模型调整时刻,确定处理第一数据的目标AI模型,初传时刻为第二数据的初始传输时刻,第二数据为目标AI模型处理第一数据后得到的数据。
在一种实施方式中,第二确定单元903,具体用于:
若初传时刻在模型调整时刻之前,则将第一AI模型确定为目标AI模型;
若初传时刻在模型调整时刻之后,则将第二AI模型确定为目标AI模型。
图9C为本公开实施例提供的一种模型同步装置的结构示意图三。如图9C所示,装置还包括接收单元904,接收单元904用于:
接收第二设备或者第三方管理设备发送的第二AI模型,第二AI模型的接收时刻在模型调整时刻之前或者在模型调整时刻的确定时刻之前。
图9D为本公开实施例提供的一种模型同步装置的结构示意图四。如图9D所示,装置还包括发送单元905,发送单元905用于:
向第二设备发送第二AI模型。
需要说明的是,本公开实施例中对单元的划分是示意性的,仅仅为一种逻辑功能划分,实际实现时可以有另外的划分方式。另外,在本公开各个实施例中的各功能单元可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现,也可以采用软件功能单元的形式实现。
上述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时,可以存储在一个处理器可读取存储介质中。基于这样的理解,本公开的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以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中,包括若干指令用以使得一台计算机设备(可以是个人计算机,服务器,或者网络设备等)或处理器(processor)执行本公开各个实施例方法的全部或部分步骤。而前述的存储介质包括:U盘、移动硬盘、只读存储器(read-only memory,ROM)、随机存取存储器(random access memory,RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
在此需要说明的是,本公开提供的上述装置,能够实现上述方法实施例所实现的所有方法步骤,且能够达到相同的技术效果,在此不再对本实施例中与方法实施例相同的部分及有益效果进行具体赘述。
本公开实施例还提供一种处理器可读存储介质,处理器可读存储介质存储有计算机程序,计算机程序用于使处理器执行上述方法实施例任一项所述的方法。
处理器可读存储介质可以是计算机能够存取的任何可用介质或数据存储设备,包括但不限于磁性存储器(例如软盘、硬盘、磁带、磁光盘(MO)等)、光学存储器(例如CD、DVD、BD、HVD等)、以及半导体存储器(例如ROM、EPROM、EEPROM、非易失性存储器(NAND FLASH)、固态硬盘(SSD))等。
本公开实施例还提供一种计算机程序产品,包括计算机程序,计算机程序被处理器执行时实现上述方法实施例任一项所述的方法。
本领域内的技术人员应明白,本公开的实施例可提供为方法、系统、或计算机程序产 品。因此,本公开可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且,本公开可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器和光学存储器等)上实施的计算机程序产品的形式。
本公开是参照根据本公开实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机可执行指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机可执行指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器,使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些处理器可执行指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的处理器可读存储器中,使得存储在该处理器可读存储器中的指令产生包括指令装置的制造品,该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些处理器可执行指令也可装载到计算机或其他可编程数据处理设备上,使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理,从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
显然,本领域的技术人员可以对本公开进行各种改动和变型而不脱离本公开的精神和范围。这样,倘若本公开的这些修改和变型属于本公开权利要求及其等同技术的范围之内,则本公开也意图包含这些改动和变型在内。

Claims (43)

  1. 一种模型同步方法,其特征在于,应用于通信系统中的第一设备,所述通信系统还包括第二设备,所述方法包括:
    确定所述第一设备和所述第二设备同步进行模型调整的模型调整时刻;
    根据所述模型调整时刻,将第一AI模型调整为第二AI模型。
  2. 根据权利要求1所述的方法,其特征在于,所述确定模型调整时刻,包括:
    获取模型调整信息;
    根据所述模型调整信息,确定所述模型调整时刻。
  3. 根据权利要求2所述的方法,其特征在于,所述模型调整信息包括:模型同步周期和模型调整时长;
    所述根据所述模型调整信息,确定所述模型调整时刻,包括:
    根据所述模型同步周期和所述模型调整时长,确定所述模型调整时刻。
  4. 根据权利要求3所述的方法,其特征在于,所述根据所述模型同步周期和所述模型调整时长,确定所述模型调整时刻,包括:
    获取指示信息;
    根据所述指示信息、所述模型同步周期和所述模型调整时长,确定所述模型调整时刻。
  5. 根据权利要求2所述的方法,其特征在于,所述模型调整信息包括预设时刻和偏移时长;
    所述根据所述模型调整信息,确定模型调整时刻,包括:
    根据所述预设时刻和所述偏移时长,确定所述模型调整时刻。
  6. 根据权利要求2-5任一项所述的方法,其特征在于,所述模型调整信息还包括:参考时间方式和/或模型同步时间精度。
  7. 根据权利要求2-6任一项所述的方法,其特征在于,所述通信系统还包括第三方管理设备。
  8. 根据权利要求7所述的方法,其特征在于,所述模型调整信息是通过所述第一设备、所述第二设备或者所述第三方管理设备配置;
    或者,所述模型调整信息是通过所述第一设备和所述第二设备协商确定;
    或者,所述模型调整信息是通过所述第三方管理设备和所述第一设备协商确定;
    或者,所述模型调整信息是通过所述第三方管理设备和所述第二设备协商确定;
    或者,所述模型调整信息是通过所述第三方管理设备、所述第一设备和所述第二设备协商确定。
  9. 根据权利要求1-8任一项所述的方法,其特征在于,所述将第一AI模型调整为第二AI模型,包括以下任意一项:
    删除所述第一AI模型,并添加所述第二AI模型;
    将所述第一AI模型更换、更新或修改为所述第二AI模型;
    去激活所述第一AI模型,并激活所述第二AI模型;
    停用所述第一AI模型,并启用所述第二AI模型;
    启用所述第二AI模型;
    添加所述第二AI模型;
    激活所述第二AI模型。
  10. 根据权利要求1-9任一项所述的方法,其特征在于,所述方法还包括:
    根据初传时刻和所述模型调整时刻,将所述第一AI模型或所述第二AI模型确定为处理第一数据的目标AI模型,所述初传时刻为第二数据的初始传输时刻,所述第二数据为所述目标AI模型处理所述第一数据后得到的数据。
  11. 根据权利要求10所述的方法,其特征在于,所述根据初传时刻和所述模型调整时刻,将所述第一AI模型或所述第二AI模型确定为处理第一数据的目标AI模型,包括:
    若所述初传时刻在所述模型调整时刻之前,则将所述第一AI模型确定为所述目标AI模型;
    若所述初传时刻在所述模型调整时刻之后,则将所述第二AI模型确定为所述目标AI模型。
  12. 根据权利要求1-11任一项所述的方法,其特征在于,所述第一设备包括终端设备或者网络设备;第二设备包括终端设备或者网络设备。
  13. 根据权利要求7-12任一项所述的方法,其特征在于,所述方法还包括:
    接收所述第二设备或者所述第三方管理设备发送的所述第二AI模型,所述第二AI模型的接收时刻在所述模型调整时刻之前或者在所述模型调整时刻的确定时刻之前。
  14. 根据权利要求1-13任一项所述的方法,其特征在于,所述方法还包括:
    向所述第二设备发送所述第二AI模型。
  15. 一种第一设备,其特征在于,应用于通信系统,所述通信系统还包括第二设备,所述第一设备包括存储器,收发机,处理器:
    所述存储器,用于存储计算机程序;
    所述收发机,用于在所述处理器的控制下收发数据;
    所述处理器,用于读取所述存储器中的计算机程序并执行如下操作:
    确定所述第一设备和第二设备同步进行模型调整的模型调整时刻;
    根据所述模型调整时刻,将第一AI模型调整为第二AI模型。
  16. 根据权利要求15所述的设备,其特征在于,所述处理器,具体用于执行如下操作:
    获取模型调整信息;
    根据所述模型调整信息,确定所述模型调整时刻。
  17. 根据权利要求16所述的设备,其特征在于,所述模型调整信息包括:模型同步周期和模型调整时长;所述处理器,具体用于执行如下操作:
    根据所述模型同步周期和所述模型调整时长,确定所述模型调整时刻。
  18. 根据权利要求17所述的设备,其特征在于,所述处理器,具体用于执行如下操作:
    获取指示信息;
    根据所述指示信息、所述模型同步周期和所述模型调整时长,确定所述模型调整时刻。
  19. 根据权利要求16所述的设备,其特征在于,所述模型调整信息包括预设时刻和偏移时长;所述处理器,具体用于执行如下操作:
    根据所述预设时刻和所述偏移时长,确定所述模型调整时刻。
  20. 根据权利要求16-19任一项所述的设备,其特征在于,所述模型调整信息还包括:参考时间方式和/或模型同步时间精度。
  21. 根据权利要求16-20任一项所述的设备,其特征在于,所述通信系统还包括第三方管理设备。
  22. 根据权利要求21所述的设备,其特征在于,所述模型调整信息是通过所述第一设备、所述第二设备或者所述第三方管理设备配置;
    或者,所述模型调整信息是通过所述第一设备和所述第二设备协商确定;
    或者,所述模型调整信息是通过所述第三方管理设备和所述第一设备协商确定;
    或者,所述模型调整信息是通过所述第三方管理设备和所述第二设备协商确定;
    或者,所述模型调整信息是通过所述第三方管理设备、所述第一设备和所述第二设备协商确定。
  23. 根据权利要求15-22任一项所述的设备,其特征在于,所述处理器,具体用于执行如下操作中的任意一项:
    删除所述第一AI模型,并添加所述第二AI模型;
    将所述第一AI模型更换、更新或修改为所述第二AI模型;
    去激活所述第一AI模型,并激活所述第二AI模型;
    停用所述第一AI模型,并启用所述第二AI模型;
    启用所述第二AI模型;
    添加所述第二AI模型;
    激活所述第二AI模型。
  24. 根据权利要求15-23任一项所述的设备,其特征在于,所述处理器,还用于执行如下操作:
    根据初传时刻和所述模型调整时刻,将所述第一AI模型或所述第二AI模型确定为处理第一数据的目标AI模型,所述初传时刻为第二数据的初始传输时刻,所述第二数据为所述目标AI模型处理所述第一数据后得到的数据。
  25. 根据权利要求24所述的设备,其特征在于,所述处理器,具体用于执行如下操作:
    若所述初传时刻在所述模型调整时刻之前,则将所述第一AI模型确定为所述目标AI模型;
    若所述初传时刻在所述模型调整时刻之后,则将所述第二AI模型确定为所述目标AI模型。
  26. 根据权利要求15-25任一项所述的设备,其特征在于,所述第一设备包括终端设 备或者网络设备;第二设备包括终端设备或者网络设备。
  27. 根据权利要求21-26任一项所述的设备,其特征在于,所述处理器,还用于执行如下操作:
    接收所述第二设备或者所述第三方管理设备发送的所述第二AI模型,所述第二AI模型的接收时刻在所述模型调整时刻之前或者在所述模型调整时刻的确定时刻之前。
  28. 根据权利要求15-27任一项所述的设备,其特征在于,所述处理器,还用于执行如下操作:
    向所述第二设备发送所述第二AI模型。
  29. 一种模型同步装置,其特征在于,应用于通信系统中的第一设备,所述通信系统还包括第二设备,所述装置包括:
    第一确定单元,用于确定第一设备和第二设备同步进行模型调整的模型调整时刻;
    调整单元,用于根据所述模型调整时刻,将第一AI模型调整为第二AI模型。
  30. 根据权利要求29所述的装置,其特征在于,所述第一确定单元,具体用于:
    获取模型调整信息;
    根据所述模型调整信息,确定所述模型调整时刻。
  31. 根据权利要求30所述的装置,其特征在于,所述模型调整信息包括:模型同步周期和模型调整时长;所述第一确定单元,具体用于:
    根据所述模型同步周期和所述模型调整时长,确定所述模型调整时刻。
  32. 根据权利要求31所述的装置,其特征在于,所述第一确定单元,具体用于:
    获取指示信息;
    根据所述指示信息、所述模型同步周期和所述模型调整时长,确定所述模型调整时刻。
  33. 根据权利要求30所述的装置,其特征在于,所述模型调整信息包括预设时刻和偏移时长;所述第一确定单元,具体用于:
    根据所述预设时刻和所述偏移时长,确定所述模型调整时刻。
  34. 根据权利要求30-33任一项所述的装置,其特征在于,所述模型调整信息还包括:参考时间方式和/或模型同步时间精度。
  35. 根据权利要求30-34任一项所述的装置,其特征在于,所述通信系统还包括第三方管理设备。
  36. 根据权利要求35所述的装置,其特征在于,所述模型调整信息是通过所述第一设备、所述第二设备或者所述第三方管理设备配置;
    或者,所述模型调整信息是通过所述第一设备和所述第二设备协商确定;
    或者,所述模型调整信息是通过所述第三方管理设备和所述第一设备协商确定;
    或者,所述模型调整信息是通过所述第三方管理设备和所述第二设备协商确定;
    或者,所述模型调整信息是通过所述第三方管理设备、所述第一设备和所述第二设备协商确定。
  37. 根据权利要求29-36任一项所述的装置,其特征在于,所述调整单元,具体用于实现如下任意一项:
    删除所述第一AI模型,并添加所述第二AI模型;
    将所述第一AI模型更换、更新或修改为所述第二AI模型;
    去激活所述第一AI模型,并激活所述第二AI模型;
    停用所述第一AI模型,并启用所述第二AI模型;
    启用所述第二AI模型;
    添加所述第二AI模型;
    激活所述第二AI模型。
  38. 根据权利要求29-37任一项所述的装置,其特征在于,所述装置还包括:
    第二确定单元,用于根据初传时刻和所述模型调整时刻,将所述第一AI模型或所述第二AI模型确定为处理第一数据的目标AI模型,所述初传时刻为第二数据的初始传输时刻,所述第二数据为所述目标AI模型处理所述第一数据后得到的数据。
  39. 根据权利要求38所述的装置,其特征在于,所述第二确定单元,具体用于:
    若所述初传时刻在所述模型调整时刻之前,则将所述第一AI模型确定为所述目标AI模型;
    若所述初传时刻在所述模型调整时刻之后,则将所述第二AI模型确定为所述目标AI模型。
  40. 根据权利要求29-39任一项所述的装置,其特征在于,所述第一设备包括终端设备或者网络设备;第二设备包括终端设备或者网络设备。
  41. 根据权利要求35-40任一项所述的装置,其特征在于,所述装置还包括:
    接收单元,用于接收所述第二设备或者所述第三方管理设备发送的所述第二AI模型,所述第二AI模型的接收时刻在所述模型调整时刻之前或者在所述模型调整时刻的确定时刻之前。
  42. 根据权利要求29-41任一项所述的装置,其特征在于,所述装置还包括:
    向所述第二设备发送所述第二AI模型。
  43. 一种处理器可读存储介质,其特征在于,所述处理器可读存储介质存储有计算机程序,所述计算机程序用于使所述处理器执行权利要求1-14任一项所述的方法。
PCT/CN2023/137609 2022-12-16 2023-12-08 一种模型同步方法、装置、设备及存储介质 WO2024125421A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211622245.9 2022-12-16
CN202211622245.9A CN118211670A (zh) 2022-12-16 2022-12-16 一种模型同步方法、装置、设备及存储介质

Publications (1)

Publication Number Publication Date
WO2024125421A1 true WO2024125421A1 (zh) 2024-06-20

Family

ID=91449225

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/137609 WO2024125421A1 (zh) 2022-12-16 2023-12-08 一种模型同步方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN118211670A (zh)
WO (1) WO2024125421A1 (zh)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080270329A1 (en) * 2005-09-13 2008-10-30 Long Philip M Systems and methods for martingale boosting in machine learning
US20210019152A1 (en) * 2019-07-15 2021-01-21 Microsoft Technology Licensing, Llc Data parallelism in distributed training of artificial intelligence models
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
CN114143799A (zh) * 2020-09-03 2022-03-04 华为技术有限公司 通信方法及装置
CN114844785A (zh) * 2021-02-01 2022-08-02 大唐移动通信设备有限公司 通信系统中的模型更新方法、装置及存储介质
CN115238895A (zh) * 2021-04-22 2022-10-25 华为云计算技术有限公司 模型转换方法、装置及相关设备

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080270329A1 (en) * 2005-09-13 2008-10-30 Long Philip M Systems and methods for martingale boosting in machine learning
US20210019152A1 (en) * 2019-07-15 2021-01-21 Microsoft Technology Licensing, Llc Data parallelism in distributed training of artificial intelligence models
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
CN114143799A (zh) * 2020-09-03 2022-03-04 华为技术有限公司 通信方法及装置
CN114844785A (zh) * 2021-02-01 2022-08-02 大唐移动通信设备有限公司 通信系统中的模型更新方法、装置及存储介质
CN115238895A (zh) * 2021-04-22 2022-10-25 华为云计算技术有限公司 模型转换方法、装置及相关设备

Also Published As

Publication number Publication date
CN118211670A (zh) 2024-06-18

Similar Documents

Publication Publication Date Title
US10904815B2 (en) Method for transmitting configuration message and apparatus
US20240056825A1 (en) Ue capability information processing method and apparatus, device, and storage medium
US11882566B2 (en) Communications method and apparatus
US20230232289A1 (en) Timing advance compensation method, base station, terminal, and storage medium
CN114071612B (zh) 辅小区组的主小区更新方法、装置及存储介质
CN114501528B (zh) 时延抖动同步方法、装置及存储介质
CN112543450A (zh) 密钥推衍方法及装置
WO2024093656A1 (zh) 一种无线资源管理rrm测量的配置方法及装置
WO2024125421A1 (zh) 一种模型同步方法、装置、设备及存储介质
WO2024139524A1 (zh) 辅助信息上报方法及装置
WO2022206774A1 (zh) 一种prs配置信息的确定方法及装置
WO2024027626A1 (zh) QoE测量配置方法、主节点及辅节点
WO2024169562A1 (zh) 波束失败处理方法、装置、终端及网络设备
WO2023231767A1 (zh) 定时提前值传输方法、装置及存储介质
WO2024093606A1 (zh) 底层触发移动性的处理方法和装置
WO2024094220A1 (zh) 通信方法及装置
WO2024022430A1 (zh) Mdt配置方法及装置
WO2023020276A1 (zh) 组播广播业务数据传输方法、装置、设备以及存储介质
WO2024120194A1 (zh) 算网融合方法、装置及存储介质
WO2023143258A1 (zh) 数据处理方法与装置
WO2024016970A1 (zh) L1/l2移动性配置信息的处理方法、装置及存储介质
WO2024082905A1 (zh) 协议数据单元pdu会话的建立方法及装置
WO2024022188A1 (zh) 激活/去激活小区的方法、装置及存储介质
WO2024017069A1 (zh) 定位方法、装置及存储介质
WO2024027383A1 (zh) 数据处理方法及装置

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: 23902617

Country of ref document: EP

Kind code of ref document: A1