Method and apparatus for constructing basic model of semantic knowledge base, and participating device and server

Abstract

A method and apparatus for constructing a basic model of a semantic knowledge base, and a participating device and a server. The method comprises: training an initial semantic model on the basis of initial model parameters received from a server, so as to obtain intermediate model parameters; sending the intermediate model parameters to the server, so as to instruct the server to aggregate the intermediate model parameters to obtain a global semantic model; and receiving global model parameters corresponding to the global semantic model, and on the basis of the global model parameters, constructing a local semantic model. A plurality of participating devices cooperate with a server to construct a semantic model, and thus the constructed semantic model has a relatively high degree of matching with each participating device. Furthermore, each participating device can determine a local semantic model on the basis of its own feature and a global semantic model, thereby improving the degree of matching between the determined local semantic models and the participating devices. Moreover, the server determines initial model parameters on the basis of semantic information and issues the initial model parameters, thereby improving the accuracy and security of a semantic model.

Description

Method and device for constructing semantic knowledge base basic model, participating device and server

[0001] Related applications

[0002] The present application claims priority to the Chinese patent application No. 202410832281.0, filed on June 26, 2024, and entitled "Method and device for constructing semantic knowledge base basic model, participating device and server", the content of which is incorporated herein by reference in its entirety. TECHNICAL FIELD

[0003] The present application relates to the technical field of semantic communication, and in particular to a method and device for constructing a semantic knowledge base basic model, a participating device and a server. BACKGROUND

[0004] Semantic communication is a new communication architecture that integrates user needs and information meaning into the communication process. The semantic knowledge base model is a key enabling technology for semantic communication, which can provide global knowledge background for the extraction, identification, transmission, understanding and reasoning process of semantic elements in semantic communication, as well as store search services. Therefore, the semantic knowledge base model is widely used in various semantic communication fields.

[0005] In the conventional technology, the semantic knowledge base model used in the semantic communication process between communication devices is usually obtained from a server and directly used. However, the semantic model constructed on the server cannot be matched with each communication device, resulting in poor effect of semantic communication by the communication devices using the semantic knowledge base model.

[0006] SUMMARY

[0007] The embodiments of the present application provide a method and device for constructing a semantic knowledge base basic model, a participating device and a server, which can improve the matching degree of the semantic model and each participating device.

[0008] In a first aspect, the present application provides a method for constructing a semantic knowledge base basic model. The method comprises: training an initial semantic model according to initial model parameters received from a server to obtain intermediate model parameters; sending the intermediate model parameters to the server to instruct the server to aggregate the intermediate model parameters to obtain a global semantic model; receiving global model parameters corresponding to the global semantic model, and constructing a local semantic model according to the global model parameters.

[0009] In some embodiments, the constructing a local semantic model according to the global model parameters comprises: determining a local device capability; and constructing the local semantic model according to the local device capability and the global model parameters.

[0010] In some embodiments, the constructing the local semantic model according to the local device capability and the global model parameter comprises: constructing an initial local semantic model according to the global model parameter, and determining whether the device capability meets a preset first training requirement; if yes, performing personalized training on the initial local semantic model according to local sample information in a local knowledge base and the level of the device capability to obtain the local semantic model; and if not, determining the initial local semantic model as the local semantic model.

[0011] In some embodiments, the performing personalized training on the initial local semantic model according to the local sample information in the local knowledge base and the device capability to obtain the local semantic model comprises: if the level of the device capability is a first level, performing fine-grained training on the initial local semantic model according to the sample information to obtain the local semantic model; and if the level of the device capability is a second level, performing lightweight training on the initial local semantic model according to the sample information to obtain the local semantic model.

[0012] In some embodiments, the initial model parameter is determined by the server based on semantic communication information, and the semantic communication information comprises at least one of a semantic communication target, a semantic communication intention, and a semantic communication task.

[0013] In some embodiments, the initial model parameter comprises at least one of scene information, transmission context information, a transmitted signal feature, and a channel feature.

[0014] In some embodiments, the device capability comprises at least one of a computing resource, an electric quantity, a memory, and a data transmission bandwidth.

[0015] The present application provides, in a second aspect, a method for constructing a semantic knowledge base base model, comprising: sending an initial model parameter to a plurality of participating devices to instruct each of the participating devices to perform preliminary training on an initial semantic model based on the initial model parameter to obtain an intermediate model parameter; receiving the intermediate model parameter sent by each of the participating devices; performing aggregation processing on a plurality of the intermediate model parameters to obtain a global model parameter; and sending the global model parameter to each of the participating devices to instruct each of the participating devices to construct a local semantic model according to the global model parameter.

[0016] In some embodiments, the performing aggregation processing on a plurality of the intermediate model parameters to obtain a global model parameter comprises: performing aggregation processing on a plurality of the intermediate model parameters according to a preset federated learning algorithm to obtain an aggregated model parameter; and determining the global model parameter according to the aggregated model parameter.

[0017] In some embodiments, the determining the global model parameter according to the aggregated model parameter comprises: constructing a corresponding aggregated semantic model according to the aggregated model parameter; determining whether the aggregated semantic model meets a preset second training requirement, if yes, determining the aggregated model parameter as the global model parameter; if not, returning the aggregated model parameter to the plurality of participating devices for retraining.

[0018] In some embodiments, the preset second training requirement comprises any one of: convergence of the aggregated semantic model, and a preset number of training times of the initial semantic model.

[0019] In some embodiments, the initial model parameter is determined by the server based on semantic communication information, and the semantic communication information comprises at least one of a semantic communication target, a semantic communication intention, and a semantic communication task.

[0020] In some embodiments, the initial model parameter comprises at least one of scene information, transmission context information, transmitted signal features, and channel features.

[0021] The present application provides, in a third aspect, a device for constructing a semantic knowledge base basic model, comprising: a training module configured to train an initial semantic model according to an initial model parameter received from a server to obtain intermediate model parameters; a first sending module configured to send the intermediate model parameters to the server to instruct the server to aggregate the intermediate model parameters to obtain a global semantic model; and a first receiving module configured to receive global model parameters corresponding to the global semantic model, and construct a local semantic model according to the global model parameters.

[0022] The present application provides, in a fourth aspect, a device for constructing a semantic knowledge base basic model, comprising: a second sending module configured to send an initial model parameter to a plurality of participating devices to instruct each of the participating devices to preliminarily train an initial semantic model based on the initial model parameter to obtain intermediate model parameters; a second receiving module configured to receive the intermediate model parameters sent by each of the participating devices; an aggregation module configured to aggregate the plurality of intermediate model parameters to obtain global model parameters; and a third sending module configured to send the global model parameters to each of the participating devices to instruct each of the participating devices to construct a local semantic model according to the global model parameters.

[0023] The application provides a participating device in a fifth aspect, comprising a transmitter, a processor and a receiver; the processor is configured to train an initial semantic model according to initial model parameters received from a server to obtain intermediate model parameters; the transmitter is configured to send the intermediate model parameters to the server to instruct the server to aggregate the intermediate model parameters to obtain a global semantic model; and the receiver is configured to receive global model parameters corresponding to the global semantic model and construct a local semantic model according to the global model parameters.

[0024] The application provides a server in a sixth aspect, comprising a transmitter, a processor and a receiver; the transmitter is configured to send initial model parameters to a plurality of participating devices to instruct each participating device to preliminarily train an initial semantic model based on the initial model parameters to obtain intermediate model parameters; the receiver is configured to receive the intermediate model parameters sent by each participating device; the processor is configured to aggregate a plurality of the intermediate model parameters to obtain global model parameters; and the transmitter is further configured to send the global model parameters to each participating device to instruct each participating device to construct a local semantic model according to the global model parameters.

[0025] The application provides a computer readable storage medium in a seventh aspect, which stores a computer program, and the computer program is executed by a processor to implement the construction method of the semantic knowledge base basic model provided in each embodiment of the first aspect or the second aspect.

[0026] The application provides a computer program product in an eighth aspect, comprising a computer program, and the computer program is executed by a processor to implement the construction method of the semantic knowledge base basic model provided in each embodiment of the first aspect or the second aspect.

[0027] The construction method, device, participating device and server of the semantic knowledge base basic model are configured to train an initial semantic model according to initial model parameters received from a server to obtain intermediate model parameters, send the intermediate model parameters to the server to instruct the server to aggregate the intermediate model parameters to obtain a global semantic model, receive global model parameters corresponding to the global semantic model, and construct a local semantic model according to the global model parameters. A plurality of participating devices and the server cooperatively construct a semantic model, so that the matching degree of the constructed semantic model and each participating device is high; further, each participating device can determine a local semantic model based on the global semantic model according to its own characteristics, and the matching degree of the determined local semantic model and the participating device is further improved; in addition, the server determines initial model parameters according to semantic information and sends the initial model parameters, so that the accuracy and security of the semantic model are improved. BRIEF DESCRIPTION OF DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the following will briefly introduce the drawings needed to be used in the embodiments or the related art description. Obviously, the drawings in the following description only some of the embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0029] Figure 1 is an application environment diagram of the construction method of the semantic knowledge base basic model in an embodiment of the present application.

[0030] Figure 2 is a flow diagram of the construction method of the semantic knowledge base basic model in an embodiment of the present application.

[0031] Figure 3 is a flow diagram of the construction method of the semantic knowledge base basic model in another embodiment of the present application.

[0032] Figure 4 is a flow diagram of the construction method of the semantic knowledge base basic model in another embodiment of the present application.

[0033] Figure 5 is a flow diagram of the construction method of the semantic knowledge base basic model in another embodiment of the present application.

[0034] Figure 6 is a flow diagram of the construction method of the semantic knowledge base basic model in another embodiment of the present application.

[0035] Figure 7 is a flow diagram of the construction method of the semantic knowledge base basic model in another embodiment of the present application.

[0036] Figure 8 is a flow diagram of the construction method of the semantic knowledge base basic model in another embodiment of the present application.

[0037] Figure 9 is a flow diagram of the construction method of the semantic knowledge base basic model in another embodiment of the present application.

[0038] Figure 10 is a structural block diagram of the construction device of the semantic knowledge base basic model in an embodiment of the present application.

[0039] Figure 11 is a structural block diagram of the construction device of the semantic knowledge base basic model in another embodiment of the present application.

[0040] Figure 12 is an internal structure diagram of the participating device in an embodiment of the present application.

[0041] Figure 13 is an internal structure diagram of the server in an embodiment of the present application. DETAILED DESCRIPTION

[0042] In order to make the purposes, technical solutions and advantages of the present application clearer, further detailed description will be made to the present application in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0043] The method for constructing a semantic knowledge base basic model provided by the embodiments of the present application can be applied in an application environment as shown in FIG. 1. In the application environment, a participating device 102 communicates with a server 104 through a network. The participating device 102 can be a server or a terminal. The terminal can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things device can be a smart speaker, a smart television, a smart air conditioner, a smart vehicle-mounted device, a projection device, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The head-mounted device can be a virtual reality (VR) device, an augmented reality (AR) device, smart glasses, etc. The server 104 can be a stand-alone physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0044] In the conventional method, a semantic model can be constructed by network parameters, and the size of the network model is reduced by quantization. However, the conventional method will cause precision loss, resulting in semantic information mismatch. Based on the above problems, the global semantic communication knowledge base is constructed by the way of federated learning in the present application, so as to solve the problem of semantic model matching.

[0045] The technical solutions of the present application and how the technical solutions of the present application solve the above technical problems will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described again in some embodiments. The embodiments of the present application will be described below in combination with the drawings.

[0046] In an embodiment of the present application, as shown in FIG. 2, a method for constructing a semantic knowledge base basic model is provided. Taking the participating device in FIG. 1 as an example, the method comprises the following steps S201 to S203.

[0047] S201: training an initial semantic model according to initial model parameters received from a server to obtain intermediate model parameters.

[0048] Specifically, the initial model parameters are determined by the server based on semantic communication information, and the semantic communication information comprises at least one of a semantic communication target, a semantic communication intention and a semantic communication task.

[0049] In the embodiments of the present application, the server constructs an initial semantic model according to the semantic communication information, and sends initial model parameters of the initial semantic model to a plurality of participating devices. After receiving the initial model parameters, each participating device trains the initial semantic model based on pre-stored sample data according to the initial model parameters, obtains an intermediate semantic model, and the intermediate model parameters are parameters corresponding to the intermediate semantic model. Illustratively, the initial semantic model can include a control G for completing a machine operation; the initial global model parameters are constructed, wherein the parameters include: a context C, a time t at which the machine is located, a temperature T at which the machine is located, an intensity I of a signal received by the machine, a direction D of the signal received by the machine, a transmission signal frequency f used by the machine, a transmission energy Q used by the machine, and a channel bandwidth B received by the machine.

[0050] In some embodiments, the initial semantic model can be any one of a convolutional neural network (CNN), a recurrent neural network (RNN), a transformer model, a bidirectional encoder representations from transformers (BERT), a generative pre-trained transformer (GPT), and a large language model meta AI (LLaMA).

[0051] In some embodiments, the pre-stored sample data can be historical data of semantic communication between the participating device and other devices, or sample data pre-acquired by the participating device from a third-party device.

[0052] In some embodiments, the initial model parameters include at least one of scene information, transmission context information, signal characteristics of the transmission, and channel characteristics. The scene information includes environmental information, for example, the scene information can include at least one of local time, signal strength, signal / transmission direction, temperature, humidity. The transmission context information is semantic information parsed by the terminal in the communication process, for example, the transmission context information can include at least one of no reasonable semantic information, incomplete semantic information and consistent with the target intent, incomplete semantic information and inconsistent with the target intent, complete semantic information but the transmission information is known information, complete semantic information and inconsistent with the target intent, complete semantic information. The signal characteristics are the signal conditions received or transmitted by the device, for example, the signal characteristics can include at least one of amplitude, period, phase, frequency spectrum, Fourier technique, convolution, modulation, energy, power. The channel characteristics are the channel conditions used by the device to receive or transmit signals, for example, the channel characteristics include at least one of bandwidth, noise, semantic timeliness, symbol rate, data rate, semantic similarity, semantic value.

[0053] In some embodiments, the intermediate model parameters can be the parameter gradient values between the initial model parameters and the trained model parameters. For example, the participating device trains the local model based on the communication information such as time, temperature, strength of received signal, direction of received signal, received channel bandwidth, etc. obtained in this interaction, updates the local model parameters {G, t i , T i , I i , D i , f i , Q i , B i}, where i can be the current iteration number; and updates the parsed received data semantic information condition C i into the transmission context information; further, the participating device determines the parameter gradient values between the local model parameters and the initial model parameters, which can be represented as {G, ΔC i , Δt i , ΔT i , ΔI i , ΔD i , Δf i , ΔQ i , ΔB i}.

[0054] S202: Send the intermediate model parameters to the server to instruct the server to aggregate the intermediate model parameters to obtain a global semantic model.

[0055] In the embodiments of the present application, the participating devices send the intermediate model parameters to the server, the server receives different intermediate model parameters sent by the plurality of participating devices, aggregates the plurality of intermediate model parameters by using a preset federated learning algorithm to obtain aggregated model parameters, and determines an aggregated semantic model according to the aggregated model parameters, and determines a global semantic model according to the aggregated semantic model.

[0056] Specifically, the preset federated learning algorithm can be any one of a federated averaging algorithm (FedAvg), a federated dual algorithm (FedDual), a federated asynchronous training algorithm (FedAT), or any other federated learning algorithm.

[0057] In some embodiments, the global semantic model can be obtained by determining whether the aggregated semantic model converges and / or whether the aggregated semantic model reaches a preset number of training times. Illustratively, when the aggregated semantic model converges and / or the aggregated semantic model reaches the preset number of training times, the aggregated semantic model is determined as the global semantic model; when the aggregated semantic model does not converge and / or the aggregated semantic model does not reach the preset number of training times, the aggregated model parameters are sent to each participating device for further training.

[0058] S203: Receive global model parameters corresponding to the global semantic model, and construct a local semantic model according to the global model parameters.

[0059] In the embodiments of the present application, the server sends global model parameters corresponding to the global semantic model to each participating device, and each participating device constructs a global semantic model according to the global model parameters, so as to determine a local semantic model according to the global semantic model. It should be noted that the semantic model in the embodiments of the present application is a semantic knowledge base base model.

[0060] Specifically, each participating device can directly use the global semantic model as the local semantic model, or each participating device can optimize the global semantic model according to its own characteristics to obtain the local semantic model.

[0061] In the method for constructing the semantic knowledge base base model, the initial semantic model is trained according to the initial model parameters received from the server to obtain intermediate model parameters, and the intermediate model parameters are sent to the server to instruct the server to aggregate the intermediate model parameters to obtain a global semantic model; further, the global model parameters corresponding to the global semantic model are received, and the local semantic model is constructed according to the global model parameters. Multiple participating devices and the server cooperatively construct the semantic model, so that the constructed semantic model has a higher matching degree with each participating device; further, each participating device can determine the local semantic model based on the global semantic model according to its own characteristics, thereby improving the matching degree of the determined local semantic model with the participating device; and the server determines the initial model parameters according to the semantic information and issues them, thereby improving the accuracy and security of the semantic model.

[0062] In one embodiment, as shown in FIG. 3, the step of constructing the local semantic model according to the global model parameters can further include steps S301-S302.

[0063] S301: Determine the local device capability.

[0064] In the embodiment of the present application, the participating device can determine the local device capability according to the local hardware configuration information. Specifically, the device capability includes at least one of computing power resources, power, memory, and data transmission bandwidth.

[0065] S302: Construct the local semantic model according to the local device capability and the global model parameters.

[0066] In the embodiment of the present application, it is determined whether the global semantic model corresponding to the global model parameters is used as the local semantic model according to the local device capability.

[0067] As a feasible implementation, the local evaluation value can be obtained according to the evaluation of the local device capability, if the evaluation value is greater than or equal to the preset threshold, the local semantic model is constructed according to the global model parameters; if the evaluation value is less than the preset threshold, the global model parameters are updated, and the local semantic model is constructed according to the updated model parameters.

[0068] In the embodiment, the local semantic model is constructed based on the global model parameters according to the local device capability, so that the local semantic model not only matches the local communication information, but also matches the local device capability.

[0069] In one embodiment, as shown in FIG. 4, the step of constructing the local semantic model according to the local device capability and the global model parameters can further include steps S401-S403.

[0070] S401: Construct an initial local semantic model according to the global model parameters, and determine whether the local device capability meets the preset first training requirement.

[0071] The first training requirement can be information such as computing resource, power, memory required for personalized training of the initial local semantic model.

[0072] In the embodiment of the application, the first training requirement and the correspondence between each requirement in the first training requirement and each capability of the device capability are stored in advance, and the threshold of each requirement in the first training requirement is determined. Further, according to the global model parameters, the initial local semantic model is constructed, and it is determined whether the capability value of the local device capability corresponding to each requirement is greater than the threshold corresponding to each requirement. If the capability value is not less than the corresponding threshold, it is determined that the device capability meets the preset first training requirement. If the capability value is less than the corresponding threshold, it is determined that the device capability does not meet the preset first training requirement.

[0073] S402: If yes, the initial local semantic model is personalized trained to obtain the local semantic model according to the local sample information in the local knowledge base and the level of the device capability.

[0074] The local sample information can be historical communication information or confidential data stored in the participating device with a high security level.

[0075] In the embodiment of the application, if the device capability meets the preset first training requirement, the participating device determines the type of personalized training according to the level of the device capability, and then performs personalized training on the initial local semantic model according to the local sample information and the type of personalized training, and updates, deletes or adds the parameters in the initial local semantic model to obtain the local semantic model.

[0076] S403: If no, the initial local semantic model is determined as the local semantic model.

[0077] In the embodiment of the application, if the local device capability does not meet the preset first training requirement, it means that the participating device cannot perform personalized training on the initial local semantic model, and the initial local semantic model is directly determined as the local semantic model.

[0078] In the embodiment, it is determined whether to perform personalized training on the initial local semantic model according to the local device capability, which avoids resource waste caused by personalized training in the case of insufficient device capability. When the device capability meets the preset first training requirement, the initial local semantic model is further personalized trained according to the level of the device capability and the local sample information, which improves the matching degree of the local semantic model and the participating device on the basis of fully utilizing the resources of the participating device.

[0079] In one embodiment, as shown in FIG. 5, the step of performing personalized training on the initial local semantic model according to the local sample information and the device capability in the local knowledge base to obtain the local semantic model can further include the following steps S501-S502.

[0080] S501: If the level of the device capability is the first level, performing fine-grained training on the initial local semantic model according to the sample information to obtain the local semantic model.

[0081] In the embodiments of the present application, if the capability values of the local device capabilities corresponding to each requirement are all greater than or equal to the corresponding threshold value, it is determined that the level of the device capability is the first level, and the initial local semantic model is trained according to all or part of the sample information, and the parameters of the initial local semantic model are updated, deleted or added according to the sample information to obtain the local semantic model.

[0082] For example, if the level of the device capability of the local participating device 1 is the first level, the location, mobility, user information, etc. of the device itself can be obtained, and the model parameters are updated until the model converges.

[0083] S502: If the level of the device capability is the second level, performing light-weight training on the initial local semantic model according to the sample information to obtain the local semantic model.

[0084] In the embodiments of the present application, if the capability values of the local device capabilities corresponding to each requirement are not all greater than or equal to the corresponding threshold value, it is determined that the level of the device capability is the second level, and the initial local semantic model is light-weight trained according to the sample information to obtain the local semantic model.

[0085] Specifically, the light-weight training includes but is not limited to quantization training, pruning processing, and compression training. For example, the sample information can be filtered, and the initial local semantic model is quantization trained according to the filtered sample information to obtain the local semantic model; or the initial local semantic model is pruned according to the sample information to obtain the local semantic model.

[0086] For example, if the participating device 2 remains fixed, the signal direction received by the device, the transmission signal frequency used by the device, and the transmission energy used by the device remain unchanged, and the memory of the participating device 2 is insufficient to store the initial local semantic model, the three types of related parameters of the signal direction received by the device, the transmission signal frequency used by the device, and the transmission energy used by the device are pruned to obtain the local semantic model. If the computing power resources of the participating device 3 cannot support fine-grained training, the initial local semantic model can be trained only for the geographical location in the sample information to obtain the local semantic model.

[0087] In this embodiment, the initial local semantic model is personalized trained according to the actual situation of the participating device, and the matching degree of the local semantic model and the participating device is improved.

[0088] In an embodiment of the present application, as shown in FIG. 6, a method for constructing a semantic knowledge base base model is provided. Taking the case of applying the method to the server in FIG. 1, the method includes the following steps S601-S604.

[0089] S601: Send initial model parameters to a plurality of participating devices to instruct each participating device to preliminarily train an initial semantic model based on the initial model parameters to obtain intermediate model parameters.

[0090] Specifically, the initial model parameters are determined by the server based on semantic communication information, wherein the semantic communication information includes at least one of a semantic communication target, a semantic communication intent, and a semantic communication task.

[0091] In an embodiment of the present application, the server constructs an initial semantic model according to the semantic communication information and sends initial semantic parameters of the initial semantic model to a plurality of participating devices. After each participating device receives the initial model parameters, it trains the initial semantic model based on pre-stored sample data according to the initial model parameters to obtain an intermediate semantic model, and the intermediate model parameters are parameters corresponding to the intermediate semantic model. Illustratively, the initial semantic model can include a control G for completing machine operation; the initial global model parameters are constructed, wherein the parameters include: context C, time t at which the machine is located, temperature T at which the machine is located, intensity I of the signal received by the machine, direction D of the signal received by the machine, frequency f of the transmission signal used by the machine, emission energy Q used by the machine, and channel bandwidth B received by the machine.

[0092] In some embodiments, the initial semantic model can be any one of a convolutional neural network (CNN), a recurrent neural network (RNN), a transformer model, a bidirectional encoder representations from transformers (BERT), a generative pre-trained transformer (GPT), and a large language model meta AI (LLaMA).

[0093] In some embodiments, the pre-stored sample data can be historical data of the participating device communicating with other devices semantically, or sample data pre-acquired by the participating device from a third-party device.

[0094] In some embodiments, the initial model parameters include at least one of scene information, transmission context information, signal characteristics of the transmission, and channel characteristics.

[0095] The scene information includes environmental information, for example, the scene information can include at least one of local time, signal strength, signal / transmission direction, temperature, and humidity. The transmission context information is information about the semantic information parsed by the terminal during the communication process, for example, the transmission context information can include at least one of no reasonable semantic information, incomplete semantic information and consistent with the target intent, incomplete semantic information and inconsistent with the target intent, complete semantic information but the transmission information is known information, complete semantic information and inconsistent with the target intent, and complete semantic information. The signal characteristics are the conditions of the signals received or transmitted by the device, for example, the signal characteristics can include at least one of amplitude, period, phase, frequency spectrum, Fourier technique, convolution, modulation, energy, and power. The channel characteristics are the conditions of the channel used by the device to receive or transmit signals, for example, the channel characteristics include at least one of bandwidth, noise, semantic timeliness, symbol rate, data rate, semantic similarity, and semantic value.

[0096] In some embodiments, the intermediate model parameters can be the parameter gradient values between the initial model parameters and the trained model parameters. For example, the participating device trains the local model based on the communication information such as the time, temperature, strength of the received signal, direction of the received signal, channel bandwidth, etc. obtained during this interaction, updates the local model parameters {G, t i , T i , I i , D i , f i , Q i , B i}, where i can be the current iteration number, and the parsed semantic information of the received data C i is updated into the transmission context information. Further, the participating device determines the parameter gradient values between the local model parameters and the initial model parameters, which can be represented as {G, ΔC i , Δt i , ΔT i , ΔI i , ΔD i , Δf i , ΔQ i , ΔB i}.

[0097] S602: receiving the intermediate model parameters sent by each participating device.

[0098] In the embodiment of the present application, the server receives the intermediate model parameters of the plurality of participating devices.

[0099] S603: aggregating the plurality of intermediate model parameters to obtain global model parameters.

[0100] In the embodiment of the present application, the participating device sends the intermediate model parameters to the server, the server receives different intermediate model parameters sent by the plurality of participating devices, aggregates the plurality of intermediate model parameters by using a preset algorithm to obtain aggregated model parameters, determines an aggregated semantic model according to the aggregated model parameters, and determines a global semantic model according to the aggregated semantic model.

[0101] S604: sending the global model parameters to each participating device to instruct each participating device to construct a local semantic model according to the global model parameters.

[0102] In the embodiment of the present application, the server sends the global model parameters corresponding to the global semantic model to each participating device, each participating device constructs an initial local semantic model according to the global model parameters, and the initial local semantic model in each participating device corresponds to the global semantic model in the server; further, each participating device determines a local semantic model according to the initial local semantic model.

[0103] In some embodiments, the participating device can directly use the initial local semantic model as the local semantic model, or the participating device can optimize the initial local semantic model according to its own characteristics to obtain the local semantic model.

[0104] In the above method for constructing a semantic knowledge base basic model, the initial model parameters are sent to a plurality of participating devices to instruct each participating device to preliminarily train an initial semantic model based on the initial model parameters to obtain intermediate model parameters; further, the server receives the intermediate model parameters sent by each participating device and aggregates the plurality of intermediate model parameters to obtain global model parameters; and the global model parameters are sent to each participating device to instruct each participating device to construct a local semantic model according to the global model parameters. The plurality of participating devices and the server cooperatively construct a semantic model, so that the constructed semantic model has a high matching degree with each participating device; further, each participating device can determine a local semantic model based on its own characteristics and the global semantic model, thereby improving the matching degree between the determined local semantic model and the participating device; in addition, the server determines the initial model parameters according to semantic information and sends them, thereby improving the accuracy and security of the semantic model.

[0105] In an embodiment of the present application, as shown in FIG. 7, the step of aggregating the plurality of intermediate model parameters to obtain the global model parameter can further include the following steps S701-S702.

[0106] S701: aggregating the plurality of intermediate model parameters according to a preset federated learning algorithm to obtain aggregated model parameters.

[0107] In an embodiment of the present application, the plurality of intermediate model parameters are aggregated according to the preset federated learning algorithm. Specifically, the aggregated model parameters can be determined from the plurality of intermediate model parameters according to the preset federated algorithm, or the aggregated model parameters can be calculated according to the preset federated algorithm and the plurality of intermediate model parameters.

[0108] In some embodiments, the preset federated learning algorithm can be any one of a federated averaging algorithm (FedAvg), a federated dual algorithm (FedDual), a federated asynchronous training algorithm (FedAT), or any other federated learning algorithm.

[0109] S702: determining the global model parameter according to the aggregated model parameter.

[0110] In an embodiment of the present application, the aggregated model parameter can be determined as the global model parameter, or an aggregated semantic model can be established according to the aggregated model parameter, the aggregated semantic model is trained to obtain a global semantic model, thereby determining the global model parameter, or the aggregated model parameter can be optimized to obtain the global model parameter.

[0111] In the present embodiment, the plurality of intermediate model parameters are aggregated according to the preset federated learning algorithm to obtain the aggregated model parameter, thereby determining the global model parameter according to the aggregated model parameter. The advantages of federated learning can improve the accuracy and security of the semantic model.

[0112] In an embodiment of the present application, as shown in FIG. 8, the step of determining the global model parameter according to the aggregated model parameter can further include the following steps S801-S802.

[0113] S801: constructing a corresponding aggregated semantic model according to the aggregated model parameter.

[0114] In an embodiment of the present application, the aggregated semantic model has the same model formula as the initial semantic model, and the aggregated model parameter is used to replace the initial model parameter in the model formula to obtain the aggregated semantic model.

[0115] S802: Determine whether the aggregated semantic model meets the preset second training requirement, if yes, determine the aggregated model parameter as the global model parameter; if not, return the aggregated model parameter to the plurality of participating devices for retraining.

[0116] Specifically, the preset second training requirement includes any of the following: (1) the aggregated semantic model converges; (2) the number of training times on the initial semantic model reaches a preset number. Illustratively, when the aggregated semantic model converges, and / or the aggregated semantic model reaches the preset number of training times, the aggregated semantic model is determined as the global semantic model; when the aggregated semantic model does not converge, and / or the aggregated semantic model does not reach the preset number of training times, the aggregated model parameter is sent to each participating device for continuous training.

[0117] In this embodiment, the global model parameter is determined according to whether the aggregated semantic model meets the preset second training requirement, thereby improving the accuracy of the global model parameter.

[0118] In an embodiment of the present application, as shown in FIG. 9, an interactive flowchart of a method for constructing a semantic knowledge base base model is provided. As shown in FIG. 9, the method includes the following steps S1-S12.

[0119] S1: The server sends the initial model parameter to the plurality of participating devices.

[0120] S2: Each participating device trains the initial semantic model according to the initial model parameter received from the server to obtain intermediate model parameters.

[0121] S3: The server receives the intermediate model parameters sent by each participating device.

[0122] S4: The server aggregates the plurality of intermediate model parameters according to a preset federated learning algorithm to obtain aggregated model parameters.

[0123] S5: The server constructs a corresponding aggregated semantic model according to the aggregated model parameters.

[0124] S6: The server determines whether the aggregated semantic model meets the preset second training requirement, if yes, determines the aggregated model parameter as the global model parameter; if not, returns the aggregated model parameter to the plurality of participating devices for retraining.

[0125] S7: The server sends the global model parameter to each participating device.

[0126] S8: Each participating device determines the local device capability.

[0127] S9: Each participating device constructs an initial local semantic model according to the global model parameter, and determines whether the device capability meets the preset first training requirement.

[0128] S10: If yes, each participating device performs S11 or S12; if no, each participating device determines the initial local semantic model as the local semantic model.

[0129] S11: If the level of the device capability of the participating device is the first level, the participating device performs fine-grained training on the initial local semantic model according to the sample information to obtain the local semantic model.

[0130] S12: If the level of the device capability of the participating device is the second level, the participating device performs lightweight training on the initial local semantic model according to the sample information to obtain the local semantic model.

[0131] In the method for constructing the semantic knowledge base base model, the initial semantic model is trained according to the initial model parameters received from the server to obtain intermediate model parameters, and the intermediate model parameters are sent to the server to instruct the server to aggregate the intermediate model parameters to obtain a global semantic model; further, global model parameters corresponding to the global semantic model are received, and a local semantic model is constructed according to the global model parameters. In this way, multiple participating devices and the server cooperatively construct a semantic model, so that the constructed semantic model has a high matching degree with each participating device; further, each participating device can determine a local semantic model based on the global model parameters according to its own characteristics, improving the matching of the determined local semantic model with the participating device; in addition, the server determines the initial model parameters according to the semantic information and issues them, improving the accuracy and security of the semantic model.

[0132] It should be understood that although each step in the flowcharts of FIGS. 2-9 is shown in sequence according to the direction of the arrow, these steps are not necessarily executed in sequence according to the direction of the arrow. Unless otherwise explicitly stated herein, there is no strict order limitation for the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in FIGS. 2-9 can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least a part of other steps or steps or stages in other steps.

[0133] In an embodiment of the present application, as shown in FIG. 10, a semantic knowledge base base model construction apparatus is provided, comprising a training module 10, a first sending module 11, and a first receiving module 12, wherein:

[0134] The training module 10 is configured to train an initial semantic model according to initial model parameters received from a server to obtain intermediate model parameters; wherein the initial model parameters are determined by the server based on semantic communication information, and the semantic communication information includes at least one of a semantic communication target, a semantic communication intention, and a semantic communication task; and the initial model parameters include at least one of scene information, transmission context information, a signal feature of transmission, and channel characteristics.

[0135] The first sending module 11 is configured to send the intermediate model parameters to the server to instruct the server to aggregate the intermediate model parameters to obtain a global semantic model.

[0136] The first receiving module 12 is configured to receive global model parameters corresponding to the global semantic model, and construct a local semantic model according to the global model parameters.

[0137] In an embodiment, the first receiving module 12 described above can include a first determining unit and a constructing unit, wherein:

[0138] The first determining unit is configured to determine a local device capability; wherein the device capability includes at least one of computing power resources, power, memory, and data transmission bandwidth.

[0139] The constructing unit is configured to construct a local semantic model according to the local device capability and the global model parameters.

[0140] In an embodiment, the constructing unit described above is specifically configured to construct an initial local semantic model according to the global model parameters, and determine whether the local device capability meets a preset first training requirement; if yes, perform individual training on the initial local semantic model according to local sample information in a local knowledge base and a level of the device capability to obtain the local semantic model; and if not, determine the initial local semantic model as the local semantic model.

[0141] In an embodiment, the constructing unit described above is specifically configured to, in a case where the level of the device capability is a first level, perform fine-grained training on the global semantic model according to the sample information to obtain the local semantic model; and in a case where the level of the device capability is a second level, perform light-weight training on the global semantic model according to the sample information to obtain the local semantic model.

[0142] In an embodiment of the present application, as shown in FIG. 11, a semantic knowledge base basic model construction device is provided, which includes a second sending module 13, a second receiving module 14, an aggregation module 15, and a third sending module 16, wherein:

[0143] The second sending module 13 is configured to send initial model parameters to the plurality of participating devices, so as to instruct the participating devices to preliminarily train the initial semantic model based on the initial model parameters to obtain intermediate model parameters; wherein the initial model parameters are determined by the server based on semantic communication information, and the semantic communication information includes at least one of a semantic communication target, a semantic communication intention, and a semantic communication task; and the initial model parameters include at least one of scene information, transmission context information, a transmitted signal feature, and a channel feature.

[0144] The second receiving module 14 is configured to receive the intermediate model parameters sent by the participating devices.

[0145] The aggregation module 15 is configured to aggregate the plurality of intermediate model parameters to obtain global model parameters.

[0146] The third sending module 16 is configured to send the global model parameters to the participating devices, so as to instruct the participating devices to construct local semantic models according to the global model parameters.

[0147] In an embodiment, the aggregation module 15 can include an aggregation unit and a second determination unit.

[0148] The aggregation unit is configured to aggregate the plurality of intermediate model parameters according to a preset federated learning algorithm to obtain aggregated model parameters.

[0149] The second determination unit is configured to determine the global model parameters according to the aggregated model parameters.

[0150] In an embodiment, the second determination unit is specifically configured to: construct a corresponding aggregated semantic model according to the aggregated model parameters; determine whether the aggregated semantic model meets a preset second training requirement; if yes, determine the aggregated model parameters as the global model parameters; and if not, return the aggregated model parameters to the plurality of participating devices for retraining; wherein the preset second training requirement includes any one of the following: convergence of the aggregated semantic model, or a training number of the initial semantic model reaching a preset number.

[0151] The specific limitations of the construction device of the semantic knowledge base base model can be referred to the limitations of the construction method of the semantic knowledge base base model in the foregoing, which will not be repeated here. Each module in the construction device of the semantic knowledge base base model can be realized by software, hardware, and a combination thereof, in whole or in part. Each module can be embedded in or independent of a processor in a computer device in a hardware form, or can be stored in a memory in a computer device in a software form, so as to be called and executed by a processor to perform the operations corresponding to each module.

[0152] Fig. 12 is a structural schematic diagram of a participating device according to an embodiment of the present application. The participating device can include a receiver 31, a memory 32, a processor 33, at least one communication bus 34, and a transmitter 35. The communication bus 34 is used to realize the communication connection between the elements. The memory 32 can include a high-speed RAM memory, and can also include a non-volatile storage NVM, for example, at least one disk memory. The memory 32 can store various programs for completing various processing functions and implementing the method steps of the embodiment. In the embodiment, the transmitter 35 can be a radio frequency processing module or a baseband processing module in the participating device, and the receiver 31 can also be a radio frequency processing module or a baseband processing module in the participating device. The transmitter 35 and the receiver 31 can be integrated together to realize a transceiver. The transmitter 35 and the receiver 31 can be coupled to the processor 33, and can realize the receiving or transmitting action under the indication or control action of the processor 33.

[0153] In the embodiment, the processor 33 is configured to train the initial semantic model according to the initial model parameter received from the server to obtain intermediate model parameters; wherein the initial model parameter is determined by the server based on semantic communication information, and the semantic communication information includes at least one of a semantic communication target, a semantic communication intention, and a semantic communication task; and the initial model parameter includes at least one of scene information, transmission context information, transmitted signal characteristics, and channel characteristics.

[0154] The transmitter 35 is configured to send the intermediate model parameters to the server to instruct the server to aggregate the intermediate model parameters to obtain a global semantic model.

[0155] The receiver 31 is configured to receive global model parameters corresponding to the global semantic model, and construct a local semantic model according to the global model parameters.

[0156] In one embodiment, the processor 33 is specifically configured to: determine a local device capability; and construct a local semantic model according to the local device capability and the global model parameters; wherein the device capability includes at least one of computing power resources, power, memory, and data transmission bandwidth.

[0157] In one embodiment, the processor 33 is specifically configured to: construct an initial local semantic model according to the global model parameters, and determine whether the device capability meets a preset first training requirement; if yes, perform individual training on the initial local semantic model according to local sample information in a local knowledge base and a level of the device capability to obtain the local semantic model; and if not, determine the initial local semantic model as the local semantic model.

[0158] In an embodiment, the processor 33 is specifically configured to: if the level of the device capability is the first level, perform fine-grained training on the initial local semantic model according to the sample information to obtain the local semantic model; and if the level of the device capability is the second level, perform lightweight training on the initial local semantic model according to the sample information to obtain the local semantic model.

[0159] FIG. 13 is a structural schematic diagram of a server provided by an embodiment of the present application. The server can include a receiver 41, a memory 42, a processor 43, at least one communication bus 44, and a transmitter 45. The communication bus 44 is used to realize the communication connection between the elements. The memory 42 can contain a high-speed RAM memory and can also include a non-volatile storage NVM, such as at least one disk memory. The memory 42 can store various programs for completing various processing functions and implementing the method steps of the present embodiment. In the present embodiment, the transmitter 45 can be a radio frequency processing module or a baseband processing module in the server, and the receiver 41 can also be a radio frequency processing module or a baseband processing module in the server. The transmitter 45 and the receiver 41 can be integrated together to realize a transceiver. Both the transmitter 45 and the receiver 41 can be coupled to the processor 43, and can realize the receiving or transmitting action under the indication or control action of the processor 43.

[0160] In the present embodiment, the transmitter 45 is used to send initial model parameters to a plurality of participating devices, to instruct the participating devices to perform preliminary training on the initial semantic model based on the initial model parameters to obtain intermediate model parameters. The initial model parameters are determined by the server based on semantic communication information, and the semantic communication information includes at least one of a semantic communication target, a semantic communication intention, and a semantic communication task. The initial model parameters include at least one of scene information, transmission context information, transmitted signal characteristics, and channel characteristics.

[0161] The receiver 41 is used to receive the intermediate model parameters sent by the participating devices.

[0162] The processor 43 is used to aggregate the plurality of intermediate model parameters to obtain global model parameters.

[0163] The transmitter 45 is also used to send the global model parameters to the participating devices, to instruct the participating devices to construct local semantic models according to the global model parameters.

[0164] In an embodiment, the processor 43 is specifically configured to: aggregate the plurality of intermediate model parameters according to a preset federated learning algorithm to obtain aggregated model parameters; and determine the global model parameters according to the aggregated model parameters.

[0165] In an embodiment, the processor 43 is specifically configured to: construct a corresponding aggregated semantic model according to the aggregated model parameters; determine whether the aggregated semantic model meets a preset second training requirement, and if so, determine the aggregated model parameters as the global model parameters; and if not, return the aggregated model parameters to the plurality of participating devices for retraining; wherein the preset second training requirement includes any one of: convergence of the aggregated semantic model; or a preset number of training times of the initial semantic model.

[0166] In an embodiment of the present application, a computer readable storage medium is provided, which stores a computer program. The computer program is executed by a processor to implement the following steps:

[0167] training the initial semantic model according to the initial model parameters received from the server to obtain intermediate model parameters; wherein the initial model parameters are determined by the server based on semantic communication information, and the semantic communication information includes at least one of a semantic communication target, a semantic communication intention, and a semantic communication task; and the initial model parameters include at least one of scene information, transmission context information, transmitted signal features, and channel features;

[0168] sending the intermediate model parameters to the server to instruct the server to aggregate the intermediate model parameters to obtain a global semantic model;

[0169] receiving global model parameters corresponding to the global semantic model, and constructing a local semantic model according to the global model parameters.

[0170] In an embodiment, the computer program is executed by the processor to further implement the following steps:

[0171] determining a device capability of the local device; wherein the device capability includes at least one of computing power resources, power, memory, and data transmission bandwidth;

[0172] constructing the local semantic model according to the device capability of the local device and the global model parameters.

[0173] In an embodiment, the computer program is executed by the processor to further implement the following steps:

[0174] constructing an initial local semantic model according to the global model parameters, and determining whether the device capability meets a preset first training requirement;

[0175] if so, performing individual training on the initial local semantic model according to local sample information in a local knowledge base and a level of the device capability to obtain the local semantic model;

[0176] if not, determining the initial local semantic model as the local semantic model.

[0177] In an embodiment, the computer program, when executed by the processor, further implements the following steps:

[0178] If the level of the device capability is the first level, the initial local semantic model is fine-tuned according to the sample information to obtain the local semantic model.

[0179] If the level of the device capability is the second level, the initial local semantic model is lightened according to the sample information to obtain the local semantic model.

[0180] In an embodiment, the computer program, when executed by the processor, further implements the following steps:

[0181] The initial model parameters are sent to the plurality of participating devices to instruct the participating devices to preliminarily train the initial semantic model based on the initial model parameters to obtain intermediate model parameters; wherein the initial model parameters are determined by the server based on semantic communication information, and the semantic communication information includes at least one of a semantic communication target, a semantic communication intention, and a semantic communication task; and the initial model parameters include at least one of scene information, transmission context information, transmitted signal characteristics, and channel characteristics;

[0182] The intermediate model parameters sent by the participating devices are received;

[0183] The plurality of intermediate model parameters are aggregated to obtain global model parameters;

[0184] The global model parameters are sent to the participating devices to instruct the participating devices to construct local semantic models according to the global model parameters.

[0185] In an embodiment, the computer program, when executed by the processor, further implements the following steps: the plurality of intermediate model parameters are aggregated according to a preset federated learning algorithm to obtain aggregated model parameters;

[0186] The global model parameters are determined according to the aggregated model parameters.

[0187] In an embodiment, the computer program, when executed by the processor, further implements the following steps: a corresponding aggregated semantic model is constructed according to the aggregated model parameters;

[0188] It is determined whether the aggregated semantic model meets a preset second training requirement; if yes, the aggregated model parameters are determined as the global model parameters; and if no, the aggregated model parameters are returned to the plurality of participating devices for retraining;

[0189] The preset second training requirement includes any one of the following: the aggregated semantic model converges; and the number of times of training of the initial semantic model reaches a preset number of times.

[0190] The embodiment of the application further provides a computer program product comprising a computer program, which, when executed by a processor, causes the processor to perform the following steps:

[0191] training the initial semantic model according to initial model parameters received from the server, to obtain intermediate model parameters; wherein the initial model parameters are determined by the server based on semantic communication information, the semantic communication information comprising at least one of a semantic communication target, a semantic communication intention, and a semantic communication task; and the initial model parameters comprising at least one of scene information, transmission context information, a signal feature of transmission, and channel characteristics;

[0192] sending the intermediate model parameters to the server, to instruct the server to aggregate the intermediate model parameters to obtain a global semantic model;

[0193] receiving global model parameters corresponding to the global semantic model, and constructing a local semantic model according to the global model parameters.

[0194] In one embodiment, the computer program, when executed by the processor, further causes the processor to perform the following steps:

[0195] determining a local device capability; wherein the device capability comprises at least one of computing power resources, power, memory, and data transmission bandwidth;

[0196] constructing the local semantic model according to the local device capability and the global model parameters.

[0197] In one embodiment, the computer program, when executed by the processor, further causes the processor to perform the following steps:

[0198] constructing an initial local semantic model according to the global model parameters, and determining whether the device capability meets a preset first training requirement;

[0199] if yes, performing individual training on the initial local semantic model according to local sample information in a local knowledge base and a level of the device capability, to obtain the local semantic model;

[0200] if no, determining the initial local semantic model as the local semantic model.

[0201] In one embodiment, the computer program, when executed by the processor, further causes the processor to perform the following steps:

[0202] if the level of the device capability is a first level, performing fine-grained training on the global semantic model according to the sample information, to obtain the local semantic model;

[0203] if the level of the device capability is a second level, performing lightweight training on the global semantic model according to the sample information, to obtain the local semantic model.

[0204] In an embodiment, the computer program, when executed by the processor, further causes the processor to perform the following steps:

[0205] sending the initial model parameters to the plurality of participating devices to instruct the participating devices to preliminarily train the initial semantic model based on the initial model parameters to obtain intermediate model parameters; wherein the initial model parameters are determined by the server based on semantic communication information, the semantic communication information including at least one of a semantic communication target, a semantic communication intention, and a semantic communication task; and the initial model parameters including at least one of scene information, transmission context information, a transmitted signal feature, and a channel feature;

[0206] receiving the intermediate model parameters sent by the participating devices;

[0207] performing aggregation processing on the plurality of intermediate model parameters to obtain global model parameters;

[0208] sending the global model parameters to the participating devices to instruct the participating devices to construct a local semantic model according to the global model parameters.

[0209] In an embodiment, the computer program, when executed by the processor, further causes the processor to perform the following steps: performing aggregation processing on the plurality of intermediate model parameters according to a preset federated learning algorithm to obtain aggregated model parameters;

[0210] determining the global model parameters according to the aggregated model parameters.

[0211] In an embodiment, the computer program, when executed by the processor, further causes the processor to perform the following steps: constructing a corresponding aggregated semantic model according to the aggregated model parameters;

[0212] determining whether the aggregated semantic model meets a preset second training requirement, and if so, determining the aggregated model parameters as the global model parameters; and if not, returning the aggregated model parameters to the plurality of participating devices for retraining;

[0213] The preset second training requirement includes any one of the following: the aggregated semantic model converges; and the number of times of training on the initial semantic model reaches a preset number of times.

[0214] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0215] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0216] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be pointed out that for those skilled in the art, without departing from the concept of the present application, some modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for constructing a basic model of a semantic knowledge base, comprising: The initial semantic model is trained based on the initial model parameters received from the server to obtain intermediate model parameters; The intermediate model parameters are sent to the server to instruct the server to aggregate the intermediate model parameters to obtain a global semantic model; Receive the global model parameters corresponding to the global semantic model, and construct a local semantic model based on the global model parameters.

2. The method according to claim 1, wherein constructing a local semantic model based on the global model parameters comprises: Determine local device capabilities; The local semantic model is constructed based on the local device capabilities and the global model parameters.

3. The method according to claim 2, wherein constructing the local semantic model based on the local device capabilities and the global model parameters comprises: An initial local semantic model is constructed based on the global model parameters, and it is determined whether the local device capabilities meet the preset first training requirements. If satisfied, the initial local semantic model is then trained in a personalized manner based on the local sample information in the local knowledge base and the level of the local device capabilities to obtain the local semantic model. If the conditions are not met, then the initial local semantic model will be determined as the local semantic model.

4. The method according to claim 3, wherein the process of performing personalized training on the initial local semantic model to obtain the local semantic model based on local sample information in the local knowledge base and the local device capabilities comprises: If the local device capability level is Level 1, then the initial local semantic model is refined and trained based on the sample information to obtain the local semantic model; or If the local device capability level is Level 2, then the initial local semantic model is trained in a lightweight manner based on the sample information to obtain the local semantic model.

5. The method according to any one of claims 1-4, wherein the initial model parameters are determined by the server based on semantic communication information, and the semantic communication information includes at least one of semantic communication target, semantic communication intent, and semantic communication task.

6. The method according to any one of claims 1-4, wherein the initial model parameters include at least one of scene information, transmission context information, transmission signal characteristics, and channel characteristics.

7. The method according to any one of claims 2-4, wherein the local device capabilities include at least one of computing resources, power, memory, and data transmission bandwidth.

8. A method for constructing a basic model of a semantic knowledge base, comprising: The initial model parameters are sent to multiple participating devices to instruct each participating device to perform preliminary training on the initial semantic model based on the initial model parameters, so as to obtain intermediate model parameters. Receive intermediate model parameters sent by each of the participating devices; The intermediate model parameters are aggregated to obtain the global model parameters; The global model parameters are sent to each of the participating devices to instruct each of the participating devices to construct a local semantic model based on the global model parameters.

9. The method according to claim 8, wherein the aggregation of the plurality of intermediate model parameters to obtain global model parameters includes: The multiple intermediate model parameters are aggregated according to a preset federated learning algorithm to obtain aggregated model parameters; The global model parameters are determined based on the aggregated model parameters.

10. The method of claim 9, wherein determining the global model parameters based on the aggregated model parameters comprises: Construct the corresponding aggregation semantic model based on the aggregation model parameters; Determine whether the aggregated semantic model meets the preset second training requirement; If the conditions are met, the aggregated model parameters are determined as the global model parameters; if the conditions are not met, the aggregated model parameters are returned to the multiple participating devices for retraining.

11. The method of claim 10, wherein the preset second training requirement includes any one of the following: The aggregate semantic model converges; The initial semantic model has been trained a preset number of times.

12. The method according to any one of claims 8-11, wherein the initial model parameters are determined by the server based on semantic communication information, the semantic communication information including at least one of semantic communication target, semantic communication intent, and semantic communication task.

13. The method according to any one of claims 8-11, wherein the initial model parameters include at least one of scene information, transmission context information, transmission signal characteristics, and channel characteristics.

14. A device for constructing a semantic knowledge base basic model, comprising: The training module is used to train the initial semantic model based on the initial model parameters received from the server to obtain intermediate model parameters; The first sending module is used to send the intermediate model parameters to the server, so as to instruct the server to aggregate the intermediate model parameters to obtain a global semantic model; The first receiving module is used to receive the global model parameters corresponding to the global semantic model, and construct a local semantic model based on the global model parameters.

15. A device for constructing a semantic knowledge base basic model, comprising: The second sending module is used to send the initial model parameters to multiple participating devices, so as to instruct each of the participating devices to perform preliminary training on the initial semantic model based on the initial model parameters to obtain intermediate model parameters; The second receiving module is used to receive intermediate model parameters sent by each of the participating devices; The aggregation module is used to aggregate multiple intermediate model parameters to obtain global model parameters; The third sending module is used to send the global model parameters to each of the participating devices, so as to instruct each of the participating devices to construct a local semantic model based on the global model parameters.

16. An participating device, comprising: Transmitter, processor, and receiver; The processor is used to train an initial semantic model based on initial model parameters received from the server to obtain intermediate model parameters; The transmitter is used to send the intermediate model parameters to the server, so as to instruct the server to aggregate the intermediate model parameters to obtain a global semantic model; The receiver is used to receive the global model parameters corresponding to the global semantic model, and to construct a local semantic model based on the global model parameters.

17. A server, comprising: Transmitter, processor, and receiver; The transmitter is used to send initial model parameters to multiple participating devices, so as to instruct each participating device to perform preliminary training on the initial semantic model based on the initial model parameters to obtain intermediate model parameters; The receiver is used to receive intermediate model parameters sent by each of the participating devices; The processor is used to aggregate multiple intermediate model parameters to obtain global model parameters; The transmitter is further configured to send the global model parameters to each of the participating devices, so as to instruct each of the participating devices to construct a local semantic model based on the global model parameters.

18. A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method according to any one of claims 1 to 13.

19. A computer program product comprising a computer program, wherein the computer program, when executed by a processor, implements the steps of the method according to any one of claims 1 to 13.

Images