CN116595439A - A semantic communication method and computer equipment - Google Patents

Abstract

This application proposes a semantic communication method and computer equipment, which can send semantic model updates including the task category of the pragmatic task and the new modal category in the case of obtaining a new modal category of the input source of the pragmatic task Request, receive the first slice, that is, the part of the first semantic model corresponding to the task category and the new modality category, so as to use the first slice to directly obtain the second semantic model, so as to use the second semantic model to The new modal class of input sources performs the above pragmatic tasks.

Description

A semantic communication method and computer equipment

technical field

This application mainly relates to the application field of artificial intelligence, and more specifically relates to a semantic communication method and computer equipment.

Background technique

Semantic communication is a new architecture that can integrate user needs and information meaning into the communication process. This architecture is expected to become the basic paradigm of the future intelligent network of all things, and fundamentally solve the cross-system and cross-system problems existing in traditional data-based communication protocols. protocol, cross-network, cross-man-machine incompatibility and difficult intercommunication, and truly realize the transparent and intelligent communication of all things.

Among them, in the practical application of semantic communication, it is usually necessary to train the corresponding semantic model for the respective data sets of multi-modal categories of specific pragmatic tasks, and deploy it to the node of the communication network, so that the node can use the multiple A semantic model to meet the semantic processing requirements of multi-modal data. But this will undoubtedly increase the storage overhead and memory capacity requirements of the nodes, and will also cause a huge overhead of communication resources, making it impossible to popularize applications on a large scale.

Contents of the invention

In order to solve the above problems, the application provides the following technical solutions:

On the one hand, the present application proposes a semantic communication method, the method comprising:

Obtain new modal categories of input sources for pragmatic tasks;

sending a semantic model update request; the semantic model update request includes the task category of the pragmatic task and the new modality category;

receiving a first slice; the first slice is a part of a first semantic model corresponding to the task category and the new modality category;

Using the first slice, a second semantic model is obtained, so as to perform the pragmatic task on the input information source through the second semantic model.

Optionally, before the obtaining of the new modal category of the input source of the pragmatic task, the method further includes:

Obtain the task category of any pragmatic task to be executed, and the modality category of the input source of the pragmatic task to be processed;

Send a semantic model acquisition request; the semantic model acquisition request includes the task category and the modal category;

receiving a third semantic model; the third semantic model is a trained semantic model corresponding to the task category and the modality category;

Executing the third semantic model to perform the pragmatic task on the input information source to be processed through the third semantic model.

Optionally, the obtaining a second semantic model by using the first slice includes:

Using the first slice to replace the third slice in the third semantic model to obtain the second semantic model; the third semantic model refers to the implementation of the pragmatic task on the input source of the original modal category semantic model.

Optionally, in the case of obtaining a new modal category of a pragmatic task input source, the method further includes:

Determine at least one candidate slice that has been stored; the candidate slice is a partial semantic model trained for the pragmatic task;

Obtaining a candidate modality category corresponding to the candidate slice; the candidate modality category is a modality category for training a data set corresponding to the semantic model to which the candidate slice belongs;

Comparing the new modality category with the candidate modality category to obtain a corresponding comparison result;

Determining that the comparison result is that the new modality category is the same as any of the candidate modality categories, obtaining the candidate slice corresponding to the new modality category is determined as the first slice, and performing the using the Describe the first slice, obtain the second semantic model step;

If it is determined that the comparison result is that the new modality category is different from all the candidate modality categories, or it is determined that no candidate slice is stored, the step of sending a semantic model update request is performed.

On the one hand, the present application also proposes a semantic communication method, the method comprising:

Receiving a semantic model update request; the semantic model update request is a new modal category including an input source that requests the updated semantic model to perform a pragmatic task, and a task category of the pragmatic task;

Obtaining a first slice stored corresponding to the task category and the new modality category; the first slice is trained for the pragmatic task through a data set belonging to the new modality category Part of the first semantic model;

The first slice is sent, so that the semantic model update request end uses the first slice to obtain a second semantic model capable of performing the pragmatic task on the input information source with the new modality category.

Optionally, the method also includes:

Receiving a semantic model acquisition request; the semantic model acquisition request includes the task category of the pragmatic task to be executed by the semantic model requested to be acquired, and the modal category of the input source of the pragmatic task to be processed;

obtaining a third semantic model corresponding to the task category and the modality category;

Send the third semantic model.

Optionally, the method also includes:

receiving a plurality of semantic models; the plurality of semantic models are for pragmatic tasks of the same task category, obtained through training of data sets of different modality categories;

Performing parameter difference analysis on the multiple semantic models to obtain slices for the same pragmatic function included in the multiple semantic models;

The respective slices of the plurality of semantic models are associated with the task category and the corresponding modality category and stored.

Optionally, when any semantic model acquisition request is received, the method further includes:

Sending the cross-modal slice associated with the task category, so that the semantic model acquisition request end can determine the received cross-modal slice as a candidate slice of the pragmatic task for storage;

Wherein, the cross-modal slice refers to a slice contained in semantic models other than the requested semantic model among the plurality of semantic models stored in association with the task category.

Optionally, the method also includes:

Receiving a semantic model training request; the semantic model training request includes multiple modal categories of different input sources of the pragmatic task to be performed by the semantic model requested to be trained, and the task category of the pragmatic task;

obtaining respective datasets of the plurality of modality categories labeled for the task category;

According to the data set, train to obtain a semantic model corresponding to the task category and the corresponding modality category;

Perform parameter difference analysis on the multiple semantic models obtained through training, and obtain the slices for the same pragmatic function contained in each of the multiple semantic models;

The respective slices of the plurality of semantic models are associated with the task category and the corresponding modality category and stored.

In yet another aspect, the present application also proposes a computer device, the computer device includes a transceiver and a processor, wherein:

In the case where the computer device is configured as any edge node in the communication network, the processor is configured to implement:

Obtain new modal categories of input sources for pragmatic tasks;

sending a semantic model update request; the semantic model update request includes the task category of the pragmatic task and the new modality category;

receiving a first slice; the first slice is a part of a first semantic model corresponding to the task category and the new modality category;

Obtaining a second semantic model by using the first slice, so as to perform the pragmatic task on the input information source through the second semantic model;

Where the computer device is configured as a cloud of the communication network, the processor is configured to:

Receiving a semantic model update request; the semantic model update request is a new modal category including an input source that requests the updated semantic model to perform a pragmatic task, and a task category of the pragmatic task;

Obtaining a first slice stored corresponding to the task category and the new modality category; the first slice is trained for the pragmatic task through a data set belonging to the new modality category Part of the first semantic model;

The first slice is sent, so that the semantic model update request end uses the first slice to obtain a second semantic model capable of performing the pragmatic task on the input information source with the new modality category.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only It is an embodiment of the present application, and those skilled in the art can also obtain other drawings according to the provided drawings without creative work.

FIG. 1 is a schematic flowchart of an optional embodiment 1 of the semantic communication method proposed by the present application;

Fig. 2 is a schematic diagram of the segmentation processing process of different semantic models that realize the same pragmatic task in the semantic communication method proposed by the present application;

FIG. 3 is a schematic flowchart of an optional embodiment 2 of the semantic communication method proposed by the present application;

FIG. 4 is a schematic flowchart of an optional third embodiment of the semantic communication method proposed by the present application;

FIG. 5 is a schematic flowchart of an optional fourth embodiment of the semantic communication method proposed by the present application;

FIG. 6 is a schematic flowchart of an optional fifth embodiment of the semantic communication method proposed by the present application;

FIG. 7 is a schematic flowchart of an optional sixth embodiment of the semantic communication method proposed by the present application;

FIG. 8 is a schematic diagram of a hardware structure of an optional example of a computer device suitable for the semantic communication method proposed in this application.

Detailed ways

For the content described in the background technology section, in the semantic communication scenario of multi-modal data, in order to reduce the storage overhead and memory capacity requirements of each edge node in the communication network, and the communication resources caused by sending and receiving complete semantic models between different nodes Overhead, it is proposed to adjust the semantic model architecture for a single modality, train a semantic model suitable for multi-modal data through multi-modal data sets, unify multi-modal data in a semantic space, and realize multi-modal data In this way, each edge node can deploy such a semantic model, and can perform the same pragmatic task on different modal data.

However, in this semantic communication process, due to the poor scalability of the semantic model used for multi-modal data, the types of data modes that can be processed are limited, and there are also certain requirements for the storage and computing capabilities of edge nodes. , in this way, the configuration of edge nodes in the actual communication network is difficult to guarantee the comprehensive performance in terms of scalability, hardware cost overhead, etc., which will also limit the scope of application of this semantic communication method.

In order to further improve the above problems, this application proposes to slice the semantic model obtained through the training of the data set of each modality category to obtain cross-modal slices for the same pragmatic function, that is, to determine that the semantic model of each modality category contains The slice of the semantic model (that is, the part of the semantic model), so that after the modality of the input source of any pragmatic task is updated, the slice of the semantic model corresponding to the new modality category can be obtained directly, so as to realize the semantic model update and obtain the information used for Process the semantic model of the input data of the new modal category. Compared with transmitting the complete semantic model corresponding to the new modal category, the slice transmitted by this application is a partial semantic model, which greatly reduces the communication resource overhead, and the edge node’s semantic model Storage overhead and memory capacity requirements.

The following will clearly and completely describe the technical solutions in the embodiments of the application with reference to the drawings in the embodiments of the application. Apparently, the described embodiments are only some of the embodiments of the application, not all of them. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the scope of protection of this application.

Referring to FIG. 1 , it is a schematic flowchart of an optional embodiment 1 of the semantic communication method proposed in this application. This method can be applied to any edge node of the communication network, and the edge node can support the edge side of the cloud computing service, and can have A terminal device or server with a certain data processing capability, such as a business platform built on the communication network side close to the local terminal, can provide resources such as storage, computing, and network, and can sink some key business applications to the edge of the network to reduce network transmission. In addition to the bandwidth and delay loss caused by multi-stage forwarding, this application does not describe the system architecture of the communication network and its application principles in detail. This application can determine the device type of the edge node according to the application scenario of the communication network.

Based on this, in the semantic communication scenario of the communication network, as shown in Figure 1, the semantic communication method performed by any edge node may include but not limited to the following steps:

Step S11, obtaining a new modal category of the input source of the pragmatic task;

When edge nodes perform any pragmatic tasks (such as image semantic segmentation tasks or semantic recognition tasks, etc., this application does not limit the semantic communication scenarios), the semantic model structures used for different task categories of pragmatic tasks are often different relatively large, when the pragmatic task of the task category is executed for the first time, directly according to the task category and the modality category of the current input source to be processed, at least one semantic model trained for the pragmatic task of the task category In , the semantic model of the input data that can be in the input source to be processed is obtained, and the semantic model is deployed in the edge node. After receiving the input data from the input source, the edge node can execute the semantic model Performing the pragmatic task on the input data satisfies semantic communication requirements.

Wherein, the above-mentioned input information source may be the source of input data, such as at least one device such as data collection equipment, data transmission equipment, data storage equipment, data input equipment, or a communication server supporting a certain application service. The category of the input information source in this application There are no restrictions on the modal category of the input data provided by it, and it depends on the situation.

In the above semantic communication process, if the input source modality of the above pragmatic task is updated, for example, the modal category of the current input source for this pragmatic task is different from the modal category of the previous input source, we can get The modal category of the current input source, in order to distinguish the modal category of the previous input source, can be described as a new modal category of the pragmatic task input source.

Optionally, in the above semantic communication process, it is also possible that when the input source of the pragmatic task remains unchanged, the modality category of the current input data from the input source is different from the modality of the previous input data different categories, to determine the modal update of the input source of the pragmatic task, the modal category of the current input data (that is, the data to be processed for the pragmatic task to be executed) can be recorded as the new mode of the input source of the pragmatic task state category. Therefore, the present application does not limit the way of determining the new modal category of the pragmatic task input source in the above step S11, and it depends on the situation.

Step S12, sending a semantic model update request; the semantic model update request includes the task category of the pragmatic task and the new modality category;

Following the above analysis, in the case of determining the modal update of the input data from the input source of the pragmatic task and obtaining the corresponding new modal category, the original modal category used to process the input source of the pragmatic task The semantic model of the input data, which is no longer suitable for processing the input data of the new modality category, needs to obtain the semantic model of the input data suitable for processing the new modality category. Therefore, the edge node can send a The task category of the task and the semantic model update request of the new modality category of the input information source, so as to request to obtain the semantic model corresponding to the new modality category.

Optionally, the edge node can send a semantic model update request to the cloud of the communication network, so that the cloud can determine the corresponding first semantic model according to the task category and the new mode category included in the semantic model update request, which can be It is obtained by training the data set corresponding to the new modal category labeled with the pragmatic task, and this application does not describe in detail the training implementation process of the semantic model corresponding to the data set of each modal category.

In another possible implementation, when other edge nodes of the communication network store the first semantic model corresponding to the new modal category of the pragmatic task, the edge node can also send the semantic model to the other edge node The update request is to request to update the semantic model in the edge node that is not suitable for processing the input data of the new modality category, so as to obtain the required semantic model. Optionally, in the case where the edge node locally stores the fragments in the semantic model corresponding to the different modal categories of the pragmatic task, the edge node may send a semantic model update request to the local storage device to request to obtain a new The shards corresponding to the modal categories constitute the required semantic model.

It can be seen that, in different application scenarios, the edge node can send semantic model update requests to different objects, so as to request to update the semantic model in the edge node that is not suitable for processing the input data of the new modality category. It should be understood that when the edge node sends the semantic model update request to different objects, the communication method adopted (such as the type of communication protocol followed by the transmission of the semantic model update request) can be different, and can be determined according to the actual communication requirements. The specific implementation process is not described in detail.

Step S13, receiving a first slice; the first slice may be a part of the first semantic model corresponding to the task category and the new modality category;

As described above in relation to the technical solution of this application, if an edge node needs to process the input data of a certain modal category of a pragmatic task, the corresponding complete semantic model will be obtained, which will increase the storage overhead and memory capacity requirements of the edge node, and also It will increase the communication resource overhead when transmitting the semantic model between different edge nodes. In order to solve this technical problem, in the case of requesting to update the semantic model according to the above method, you can request to obtain a new input source suitable for processing the pragmatic task. The first slice in the first semantic model of the input data of the modal category, that is, the part of the first semantic model used to realize the pragmatic function of the input data of the new modal category, which is different from the one used to process the original In the semantic model of the input data of the modal category, a slice that realizes the same pragmatic function of the input data of the original modal category.

In this way, the edge node can receive the first slice sent by the cloud or other edge nodes or local storage devices. The storage capacity of the first slice is much smaller than the storage capacity of the entire first semantic model. The semantic model greatly reduces the consumption of data transmission resources and the storage resources of the edge nodes.

In the actual application of this application, for the segmentation process of each semantic model obtained by training, the realization process of obtaining the corresponding fragmentation can be realized by the cloud or the object that trains the semantic model. This application describes the fragmentation acquisition process of each semantic model There are no restrictions on its execution objects, and it depends on the situation.

Step S14, using the first slice to obtain a second semantic model, so as to perform a pragmatic task on the input information source of the new modality category through the second semantic model.

After the edge node obtains the first slice according to the above method, the first slice can be used to replace the third slice in the third semantic model for the same pragmatic task that has been executed in history to form the input information source for the pragmatic task The second semantic model of the new modality category, in this way, the second semantic model can be executed to perform a pragmatic task on the input data of the new modality category from the input information source, so as to meet the semantic communication requirement.

Exemplarily, as shown in Figure 2, combined with the above analysis, each semantic model can include basic parts and slices after cutting, assuming that data sets of different modal categories for the same pragmatic task are trained to obtain one-to-one semantic correspondence Model A and semantic model B, after parameter difference analysis of different semantic models (such as differences in data correlation, etc., this application does not limit the content of difference analysis, depending on the situation), semantic model A can be cut as the basis Part A and slice A, semantic model B can be cut into basic part B and slice B.

Among them, the network structure of basic part A and basic part B can be the same or basically the same, and can be used interchangeably. Slice A and slice B can be partial semantic models that realize the same pragmatic function, but the difference in model parameters between the two If the corresponding network structure and its parameters are quite different, it cannot directly replace the semantic processing of data of the same modal category. This application does not describe in detail the network structure and parameters of the basic parts and slices obtained by cutting the semantic model, and it depends on the situation.

Based on this, when the edge node determines to execute the pragmatic task for the first time, after obtaining the corresponding semantic model A according to the modal category A and task category of the input source to be processed, the semantic model A can be used to analyze the modal category A's input data performs this pragmatic task. After obtaining the modal category B of the input source of the pragmatic task, that is, the above-mentioned new modal category is modal category B, if we continue to perform the pragmatic task on the input data of the modal category B through the semantic model A, we cannot Satisfy the semantic communication requirements, therefore, the present application can obtain the slice B corresponding to the modal category B for the pragmatic task according to the method described above (that is, the above-mentioned first slice), and then, as shown in Figure 2, the slice B can form a semantic model C (that is, the second semantic model) with the basic part A in the semantic model A, that is, the slice A in the semantic model A is replaced by the slice B to obtain the semantic model C.

As analyzed above, slice A and slice B can achieve the same pragmatic function, the difference is that they are more suitable for data processing in the corresponding modal category, and the basic parts of the corresponding semantic model A and semantic model B are basically the same, so , the performance of the semantic model C composed of the above basic part A and slice B is very similar to the performance of the semantic model B including the basic part B and slice B, such as the difference between the output accuracy of semantic model C and semantic model B is less than the threshold (it can be a very small value, and the specific size is not limited), semantic model C can replace semantic model B to implement the same pragmatic task on the input data of modal category B, and meet the semantic communication requirements.

It can be seen that for edge nodes that perform pragmatic tasks, if the modal category that needs to process the input data of the input source of the pragmatic task is a new modal category, it only needs to obtain cross-modal slices according to the method described above , such as the above-mentioned slice B, constituting the second semantic model for the input data in the new modal category, such as the above-mentioned semantic model C, compared to directly obtaining the complete semantic model corresponding to the new modal category, such as the above-mentioned semantic model B. While enhancing the ability of the edge node to process multi-modal data, it greatly reduces the memory capacity requirements of the edge node and the overhead of communication resources.

Referring to FIG. 3 , it is a schematic flowchart of an optional embodiment 2 of the semantic communication method proposed in this application. This embodiment can still be described from the edge node side of the communication network. As shown in FIG. 3 , the method may include:

Step S31, obtaining the task category of any pragmatic task to be executed, and the modality category of the input source of the pragmatic task to be processed;

Step S32, sending a semantic model acquisition request; the semantic model acquisition request may include the above task category and modal category;

In practical applications, for any pragmatic task in semantic communication applications, the corresponding modality can be obtained through training in advance through data sets of different modal categories, that is, data sets of different modal categories labeled for the same pragmatic task. Semantic models of categories, after differential analysis, determine slices in each semantic model for this pragmatic task.

As in the above example, through the dataset A of modal category A and the dataset B of modal category B, the corresponding semantic model A and semantic model B for the same pragmatic task are respectively trained, and then semantic model A and semantic Model B cuts to obtain slice A in semantic model A and slice B in semantic model B, which can be associated with the corresponding modality category and stored.

It should be understood that for the above-mentioned data sets of other modal categories of pragmatic tasks, the corresponding semantic model can also be trained to obtain slices corresponding to the modal categories. The implementation process is similar, and this application does not do it one by one. detail. It can be seen that the semantic model used to realize the one-to-one correspondence between different modal categories for the same pragmatic task can be segmented to obtain slices using the same pragmatic function, which are associated with the corresponding modal categories and stored for sending to Each edge node in the communication network, according to the method described above, enables the edge node to obtain a semantic model for processing the input data of the corresponding modality category, and perform the pragmatic task on the input data of the modality category.

Moreover, when it is necessary to perform the above-mentioned pragmatic task on the data of the newly added modal category, a new semantic model can be obtained by training the data set of the newly added modal category marked for the pragmatic task according to the method described above , segment it to obtain slices that realize the above pragmatic functions, send the slices to the corresponding edge nodes, quickly obtain the semantic model for the newly added modal category, and realize the semantic processing of the input data for the newly added modal category, It can meet the needs of semantic communication and increase the scalability of the semantic model of edge nodes.

Based on the above analysis, for any edge node in the communication network, when it executes any pragmatic task for the first time, that is, if the edge node has not deployed any semantic model to realize the pragmatic task, it needs to be first The semantic model used to implement the pragmatic task is deployed in the edge node. For this, in order to obtain the semantic model that can accurately process the input data of the current input source of the pragmatic task, the task category of the pragmatic task can be obtained , and the modal category of the current input source to be processed, send a semantic model acquisition request including the task category and modal category, that is, a semantic model deployment request, such as to the cloud of the communication network or with the requested semantic model. Other edge nodes send the semantic model acquisition request, etc. This application does not limit the sending method of the semantic model acquisition request, which depends on the situation.

Step S33, receiving a third semantic model; the third semantic model may be a trained semantic model corresponding to the above-mentioned task category and modality category;

Step S34, executing the third semantic model, so as to perform pragmatic tasks on the input source to be processed through the third semantic model;

Following the above analysis, for edge nodes that have not deployed any semantic model to implement the above pragmatic tasks, the complete semantics of the input data of the input source to be processed for the precise processing of the pragmatic task can be obtained according to the method described above. model, that is, the above-mentioned third semantic model, which is deployed in the edge node, so that the edge node can execute the third semantic model, perform pragmatic tasks on the input data from the input source to be processed, and realize the process of this application Do not elaborate.

In some other embodiments proposed by this application, when an edge node implements a certain pragmatic task for the first time, the semantic model acquisition request sent may also include the task category of the pragmatic task, and any task corresponding to the task category may be received. A semantic model, after determining the modal category of the input information source of the pragmatic task, if it is different from the modal category of the data set used to train the semantic model, it is relatively new to obtain the input information source of the pragmatic task The modal category of the modal category can be obtained according to the described method to obtain the cross-modal slice, and thus obtain the semantic model of the input data of the input source for processing the pragmatic task.

Step S35, obtaining a new modal category of the input source of the pragmatic task;

Step S36, sending a semantic model update request; the semantic model update request includes the task category of the pragmatic task and the new modality category;

Step S37, receiving a first slice; the first slice may be a part of the first semantic model corresponding to the above-mentioned task category and the new modality category;

Regarding the implementation process of step S35-step S37, reference may be made to the description of the corresponding part of the above embodiment, and this embodiment will not describe in detail.

As an example, take the pragmatic task of image semantic segmentation task as an example for illustration. The semantic model for realizing image semantic segmentation can adopt but is not limited to the Deeplab network structure. Combined with the description content of the corresponding example in Figure 2 above, the above data Set A can be a cityscapes dataset (that is, a cityscape dataset), and dataset B can be a gta5 dataset (that is, a game scene dataset). The method of obtaining these two modal datasets and the data contained in this application The content is not detailed. In this embodiment, different data sets in the same information source domain can be expanded to obtain a semantic model for performing image semantic segmentation processing on data of a corresponding modality category.

Specifically, the Deeplab network (that is, a general initial semantic model) can be model-trained through the cityscapes dataset to obtain the corresponding semantic model A; the Deeplab network can be model-trained through the gta5 dataset to obtain the corresponding semantic model B, and then , it is possible to measure the network parameter difference corresponding to the two data sets through but not limited to CCA (Canonical Correlation Analysis, canonical correlation analysis), and cut the part of the model with a large difference to obtain the slice in the corresponding semantic model. The methods for realizing the segmentation of different semantic models are not limited in this application, and may be determined according to circumstances.

Afterwards, according to the above method, the slice A of the semantic model A for the gta5 dataset can be replaced with the slice B of the semantic model B obtained through the training of the cityscapes dataset, and after obtaining the semantic model C, if MIoU (MeanIntersection over Union, average Intersection and union ratio) is used as the semantic segmentation evaluation index to measure the performance index of the semantic model. Using the test data set from the gta5 data set, the semantic model C is tested for the performance of the image semantic segmentation function, namely:

The gta5 test data set is semantically segmented through the semantic model C, and the obtained MIoU score can be 0.507. In addition, the cityscapes test dataset is semantically segmented through the above semantic model A, and the obtained MIoU score can be 0.693; the gta5 test dataset is semantically segmented through the semantic model A, and the obtained MIoU score can be 0.256. Among them, the MIoU score can be the ratio between the actual value set labeled by the corresponding test data and the intersection and union of the predicted value set output by the model, or the score obtained according to certain rules. Usually, the MIoU score is higher. Higher, the better the corresponding semantic segmentation effect and the higher the performance of the corresponding semantic model. Therefore, if the MIoU score is equal to 1, the corresponding semantic segmentation information is completely restored during the semantic segmentation process of the corresponding test data through the corresponding semantic model; If the MIoU score is equal to 0, the semantic segmentation of the corresponding semantic model fails.

Through the performance test results of the above semantic model, it can be seen that the semantic communication results obtained by performing the same pragmatic task on the data of another modal category through the semantic model corresponding to a certain modal category The semantic processing performance of the data of a modality category is very poor, so, in the case of the modality update of the input source of the same pragmatic task, it is necessary to obtain a semantic model for the input source of the new modality category, in which In the acquisition process, in order to reduce the storage overhead and memory capacity requirements of edge nodes, it is not necessary to directly obtain the complete semantic model of the input source for the new modal category, but to obtain the corresponding semantic model of the new modal category that implements the same semantic model. The part of the semantic model that uses the function, that is, the part that is relatively different from the semantic model of other modal categories that achieve the same pragmatic task, is recorded as a slice.

As can be seen from the above test results, the performance of the second semantic model composed of the first slice and the basic part of the third semantic model is as good as that of the first semantic The performance of the model is similar, as the difference between the above-mentioned MIoU scores is less than the score threshold (the value is small, and the specific size depends on the situation). Sources perform pragmatic tasks and can satisfy semantic communication requirements.

Step S38, using the first slice to replace the third slice in the third semantic model to obtain the second semantic model;

Step S39, execute the second semantic model, so as to perform pragmatic tasks on the input information source of the new modal category through the second semantic model.

As analyzed above, the third semantic model refers to the semantic model that executes the pragmatic task on the input source of the original modal category. When the modal category of the input source of the pragmatic task is determined to change, the pragmatic task can be obtained. In the case of a new modal category of the input source, since the third semantic model is no longer suitable for performing pragmatic tasks on the input data of the new modal category, the application can start from the task category corresponding to the pragmatic task In the slices of different modal categories (that is, different parts of the semantic model), the first slice corresponding to the new modal category is obtained, and the cross-modal slice and the basic part in the third semantic model are directly obtained. A second semantic model is formed, that is, a semantic model suitable for performing pragmatic tasks on the input data of the new modality category, and satisfies the semantic communication requirements for the input data of the new modality category.

It can be seen that in the scenario where the edge node performs the same pragmatic task on different modal data, it is not necessary to obtain the complete semantic model corresponding to the different modal categories for the pragmatic task, but only needs to obtain the new modal category corresponding The slices, that is, cross-modal slices, replace the slices in the semantic model corresponding to the original modal category, and a complete semantic model for performing pragmatic tasks on the input data of the new modal category can be obtained, which reduces the need for edge The storage capacity of the node and the overhead of communication resources, as well as the memory capacity requirements of the edge node, facilitate the expansion of the semantic model of the edge node and meet the semantic communication requirements of efficiently processing multi-modal data on the edge node.

Referring to FIG. 4 , it is a schematic flowchart of the third optional embodiment of the semantic communication method proposed in this application. This embodiment can refine the implementation process of how the edge node obtains the first slice in the semantic communication method proposed above. Description, but not limited to the detailed implementation method described in this embodiment, as shown in Figure 4, the method may include:

Step S41, obtaining a new modal category of the input source of the pragmatic task;

Step S42, determine whether to store candidate slices for the pragmatic task, if so, enter step S43; if not, execute step S49;

In the embodiment of the present application, the above-mentioned candidate slices may be part of the semantic model trained for the pragmatic task. In combination with the description of the corresponding part of the above embodiment, for the same pragmatic task, data sets of different modal categories may be used. After training, the semantic model corresponding to the pragmatic task is obtained. After analyzing the differences between different semantic models that realize the same pragmatic task, the slices that realize the same pragmatic function in each semantic model are determined.

Optionally, after obtaining multiple semantic models and the slices included in the cloud for the same pragmatic task in the communication network, the cross-modal slices corresponding to each edge node of the communication network can be determined, that is, the semantic models other than the semantic model Slices contained in other semantic models, and then send each cross-modal slice to the corresponding edge node, the edge node can determine the received cross-modal slice as the candidate slice of the pragmatic task, and compare it with the corresponding pragmatic task The task category and the modality category of the input data that can perform the pragmatic task are associated and stored.

Step S43, obtaining the candidate modality category corresponding to the candidate slice;

Among them, the candidate modality category is the modality category of the data set used to train the semantic model to which the corresponding candidate slice belongs, and different modality identifiers can be configured to represent one-to-one corresponding modality categories. This application applies to different modality categories The expression method is not limited.

Step S44, comparing the new modal category with the candidate modal category to obtain a corresponding comparison result;

Step S45, according to the comparison result, determine whether there is any candidate modal category that is the same as the new modal category, and if so, proceed to step S46; if not, execute step S47;

Step S46, obtaining the candidate slice corresponding to the new modality category and determining it as the first slice;

Following the above analysis, when an edge node needs to obtain a cross-modal slice of a new modal category (such as the first slice above) for the input source of a pragmatic task, it can first check whether there is a candidate slice stored locally. Candidate slices are found, and it can be detected whether the cross-modal slice has been stored this time. It can be implemented according to but not limited to the comparison method described above. If the cross-modal slice is stored, it can be directly obtained from the storage device of the edge node. Read the cross-modal slice.

Step S47, sending a semantic model update request; the semantic model update request includes the task category of the pragmatic task and the new modality category;

Step S48, receiving a first slice; the first slice may be a part of the first semantic model corresponding to the above-mentioned task category and the new modality category;

The edge node does not store candidate slices, or after comparing the candidate modality category of the candidate slice stored in the edge node itself with the obtained new modality category, it is determined that the edge node does not store the candidate corresponding to the new modality category In the case of slicing, a semantic model update request can be sent to the cloud or other edge nodes in the communication network to request to obtain the first slice used to implement the pragmatic task for the input data of the new modality category. The implementation process can refer to The description of the corresponding part of the context is not described in detail in this embodiment.

Step S49, using the first slice to obtain a second semantic model, so as to perform a pragmatic task on the input information source of the new modality category through the second semantic model.

Regarding the implementation process of step S49, reference may be made to the description of the corresponding part of the above embodiment, and this embodiment will not describe it in detail.

Combined with the semantic communication method described above from the edge node side, the implementation process of the semantic communication method will be described from the cloud side of the communication network. The cloud can be a cloud server that supports cloud computing services, which is the control end of edge computing. To realize the communication management of different edge nodes in the communication network, this application does not describe the structure and basic functions of the cloud in detail.

Referring to FIG. 5 , it is a schematic flowchart of an optional embodiment 4 of the semantic communication method proposed in this application. As shown in FIG. 5 , the semantic communication method executed on the cloud may include:

Step S51, receiving a semantic model update request; the semantic model update request is a new modal category including an input source for requesting the updated semantic model to perform a pragmatic task, and a task category of the pragmatic task;

In the embodiment of the present application, after any edge node (that is, the semantic model update request end) obtains a new modal category of the input source of the pragmatic task, in order to realize the pragmatic task of the input data of the new modal category, The corresponding semantic model update request can be sent to the cloud, and the implementation process can refer to the description of the semantic communication method executed on the edge node side above, which will not be described in detail in this embodiment.

Step S52, obtaining the first slice stored corresponding to the task category and the new modality category; the first slice is part of the first semantic model trained for the pragmatic task through the data set belonging to the new modality category ;

Combined with the above-mentioned embodiment of the difference analysis of multiple semantic models of different modal categories for the same pragmatic task, the relevant description content of the respective slices of multiple semantic models is obtained, and the cloud can obtain the slices of these multiple semantic models, and store them It is associated with the task category and modal category of the corresponding pragmatic task and stored, so that after the cloud receives any semantic model update request, it can obtain the relationship between different slices and task categories and modal categories according to the The request contains the first slice associated with the task class and the new modal class.

It should be noted that, regarding the above-mentioned data sets of different modal categories for any pragmatic task, training a one-to-one corresponding semantic model, and then further segmenting to obtain the implementation process of the slice, and the execution object of the implementation process (such as the cloud and/or at least one edge node), reference may be made to but not limited to the description of the corresponding part of the context, and this embodiment will not be described in detail here.

Step S53 , sending the first slice, so that the semantic model update requester uses the first slice to obtain a second semantic model capable of performing pragmatic tasks on the input information source with the new modality category.

It can be seen that when any edge node needs to process the input data of a new modal category of the input source of the same pragmatic task, the cloud only needs to process the part of the semantic model corresponding to the new modal category, that is, the first everything Compared with sending the entire semantic model module, it greatly reduces the amount of transmitted data and reduces the consumption of communication resources. At the same time, it also reduces the storage overhead and memory usage of the transmitted data on the receiving end. The storage overhead and memory capacity requirements of edge nodes increase the scope of application.

For the edge node receiving the first slice, it can directly replace the third slice of the third semantic model for the original modality category, and obtain the second slice for performing pragmatic tasks on the input data of the new modality category. Semantic model, and the performance of the second semantic model is less different from the performance of the first semantic model obtained through the training of the data set of the new modality category. Using the second semantic model can meet the requirements for the new modality category. The semantic communication requirements of input data improve the scalability of the semantic model of edge nodes and meet the semantic communication of multi-modal data.

Referring to FIG. 6 , it is a schematic flowchart of an optional embodiment five of the semantic communication method proposed in this application. This embodiment can still be described from the cloud side. As shown in FIG. 6 , the semantic communication method may include:

Step S61, receiving a plurality of semantic models; the plurality of semantic models may be for pragmatic tasks of the same task category and obtained through training of data sets of different modal categories;

Step S62, performing parameter difference analysis on the multiple semantic models to obtain the slices for the same pragmatic function included in the multiple semantic models;

Step S63, associating the respective slices of the plurality of semantic models with task categories and corresponding modal categories and storing them;

Combined with the description of the corresponding part of the above embodiment, as shown in Figure 2, data sets of different modal categories can be labeled for the same pragmatic task, and then a semantic model for realizing the pragmatic task can be obtained through training each data set , such as parameter correlation difference analysis, determine the part with large parameter difference, and then realize the segmentation of the semantic model, and obtain slices for the same pragmatic function in different semantic models. In order to facilitate subsequent edge nodes to obtain cross-modal slices, The obtained slices of each semantic model for the same pragmatic task can be associated with the task type of the pragmatic task and the modality category to which the data set used for training belongs, and then stored. The specific storage method is not limited in this application, and can be It depends.

In some embodiments, the above-mentioned multiple semantic models that implement the same pragmatic task can be trained by the cloud according to the method described above, or can be obtained by one or more edge nodes, and/or those that can access the edge nodes The terminal device is trained, and then the trained semantic model is sent to the cloud, so that the cloud can obtain slices of multiple semantic models. Among them, when the cloud receives multiple trained semantic models sent by multiple edge nodes, different edge nodes can train semantic models corresponding to different modal categories of the same pragmatic task, avoiding resource waste caused by repeated training, and the implementation process This application does not describe in detail.

Optionally, in the case where any edge node or terminal device obtains the semantic models of multiple modal categories for pragmatic tasks, the multiple semantic models can be parameterized according to but not limited to the method described above. Difference analysis realizes the segmentation processing of multiple semantic models, obtains slices corresponding to each modality category, and then sends the multiple semantic models and the slices they contain to the cloud for storage.

Step S64, receiving a semantic model acquisition request; the semantic model acquisition request includes the task category of the pragmatic task to be executed by the semantic model requested to be acquired, and the modal category of the input source of the pragmatic task to be processed;

Step S65, obtaining a third semantic model corresponding to the task category and the modality category;

Step S66, sending the third semantic model, so that the semantic model acquisition requester executes the third semantic model, and executes the pragmatic task of the input information source to be processed;

Combined with the description of the corresponding part of the above edge node side embodiment, when any edge node needs to execute a certain pragmatic task for the first time, since no semantic model for implementing the pragmatic task is deployed in the edge node, it is necessary to obtain The semantic model of the pragmatic task can be realized. After the cloud obtains the semantic model acquisition request sent by the edge node, it can parse the semantic model acquisition request to obtain the task category of the pragmatic task to be implemented and the pending processing of the pragmatic task. Input the modal category of the information source, and then, from the stored semantic models of different modal categories for the pragmatic task, select the semantic model corresponding to the modal category of the input source to be processed, denoted as the first Three semantic models are sent to the corresponding edge nodes (that is, the semantic model acquisition request end) for deployment.

Optionally, in order for the terminal device to perform pragmatic tasks on the input source to be processed through other edge nodes, the cloud can also send the obtained third semantic model to each edge node in the communication network for deployment. In this way, the terminal The device sends data of the same modality category for the pragmatic task to any edge node, and the edge node can perform the pragmatic task on the data through the third semantic model.

In some other embodiments, after the cloud stores multiple slices of different modal categories for the same pragmatic task according to the above method, the modal category of the cross-modal slice of each edge node can be determined, that is, the edge node has deployed semantic Other modal categories other than the modal category corresponding to the model, after that, you can send the cross-modal slices stored in the task category association of the pragmatic task (that is, in the multiple semantic models stored in the task category association storage of the pragmatic task, Slices included in the semantic model other than the requested semantic model), such as sending the cross-modal slice to the corresponding edge node as a candidate slice for storage.

Step S67, receiving a semantic model update request;

Following the above analysis, for the above semantic model acquisition request end or other edge nodes, determine the modal update of the input source of the pragmatic task, and after obtaining the new modal category, you can send a semantic model update request to the cloud, or determine the local unavailable In the case of storing the candidate slice corresponding to the new modal category, send the semantic model update request to the cloud to request to obtain the slice corresponding to the new modal category that realizes the pragmatic task. There are no restrictions on the sending object and its triggering conditions, and it depends on the situation.

Step S68, obtaining the first slice stored corresponding to the task category of the pragmatic task and the new modality category of the input information source;

Step S69, sending the first slice to the semantic model update request end, so as to use the first slice to obtain a second semantic model capable of performing pragmatic tasks on the input source with the new modality category.

Through the content contained in the semantic model update request, after the cloud query finds that the semantic model update request end (such as any edge node, etc.) performs the first slice of the pragmatic task on the input source of the new modality category, it only needs to add the second All slices are fed back to the requester for updating the semantic model, so that it can obtain the second semantic model according to the method described above, and realize the semantic processing of the new modal data.

Since the first slice is part of the first pragmatic model, its data volume is much smaller than the data volume of the entire first semantic model. In this way, compared with sending the complete first semantic model, the cloud sends the first slice to the edge node. The consumption of communication resources is greatly reduced, and storage overhead and memory capacity requirements are also reduced for edge nodes receiving the first slice.

In some other embodiments proposed by this application, combined with the semantic communication methods described in the above embodiments, before the first node in the communication network trains the semantic model for realizing the pragmatic task, it can query whether other edge nodes and the cloud have been trained The semantic model for realizing the pragmatic task, and the modal category to which the data set used for training the trained semantic model belongs, etc., can be realized by sending a semantic model query request to determine the second node of the communication network (that is, different from the first At least one edge node of the node) or a certain number of semantic models that have been trained by the cloud server to realize the pragmatic task, the semantic model training information containing the modal category corresponding to the trained semantic model can be fed back to the first node, to This enables the first node to train semantic models of other modal categories, and then sends them to the cloud, and the cloud determines the slices included in each semantic model according to the above method.

It is determined that the second node or the cloud has trained multiple semantic models for all the modal categories of the pragmatic task, that is, the semantic models for the preset number of modal categories for the pragmatic task have been trained, that is, it is determined that the received The number of semantic models is equal to the number of preset modal categories, and the semantic model training notification information can be sent to the first node to notify the first node that there is no need to retrain the semantic model to realize the pragmatic task. According to the method described above, Get the desired semantic model directly.

Among them, for the edge nodes or terminal devices that have obtained the semantic model after training, the semantic model can be directly reported to the cloud for segmentation processing, or the corresponding slices can be obtained after segmentation, and the semantic model and its slices can be sent to the cloud for storage. This application There are no restrictions on the training implementation process of multiple semantic models corresponding to different modal categories of the same pragmatic task, and the execution objects for performing the training and even semantic model segmentation processing, which can be determined according to actual scene requirements.

Based on the above analysis, referring to FIG. 7 , it is a schematic flow diagram of an optional embodiment six of the semantic communication method proposed by this application. This embodiment can describe the training implementation process of the semantic model and its segmentation process. This method can be used in the cloud Execution, as shown in Figure 7, the semantic communication method may include:

Step S71, receiving a semantic model training request;

As analyzed above, the semantic model training request may include multiple modal categories of different input sources of the pragmatic task to be performed by the semantic model requested to be trained, and the task category of the pragmatic task, and the semantic model training request may be composed of Any edge node sends, this application does not limit its generation or trigger sending method.

Step S72, obtaining respective datasets of multiple modal categories marked for the task category;

Step S73, according to different data sets, train to obtain the semantic model corresponding to the task category and the corresponding mode category;

Combined with the training process shown in Figure 2, for datasets A and B marked with the same pragmatic task, or even more datasets, the initial semantic model can be trained and learned through the training data in the same dataset, and the useful The semantic model of the pragmatic task is implemented for the data of the corresponding modal category, such as the semantic model A corresponding to the data set A, and the semantic model B corresponding to the data set B. The training and implementation process will not be described in detail.

Step S74, performing parameter difference analysis on the multiple semantic models obtained through training, and obtaining the slices for the same pragmatic function included in the multiple semantic models;

Step S75 , associating the respective slices of the multiple semantic models with the task category and the corresponding modality category and storing them.

Regarding the implementation process of step S74 and step S75, as well as the implementation process of updating the existing semantic model in the edge node and obtaining the semantic model used to process the input data of the new modal category according to the query cross-modal slice, you can refer to the above The description of the corresponding part of the embodiment of the text is not described in detail in this embodiment.

Optionally, after obtaining the data set of the new modal category for the pragmatic task, the corresponding semantic model can still be trained according to the scheme described above, and the difference analysis between it and other trained semantic models can be obtained. The slices contained in the newly trained semantic model are added to the storage space of the cloud, so that the edge nodes can support the execution of pragmatic tasks on the data of the new modal category, and the scalability of the semantic model of the edge nodes is improved.

Optionally, this application also proposes a semantic communication device, which can be applied to any edge node of a communication network, and the device can include:

A modal category obtaining module, used to obtain a new modal category of the input source of the pragmatic task;

A semantic model update request sending module, configured to send a semantic model update request; the semantic model update request includes the task category of the pragmatic task and the new modality category;

A first slice receiving module, configured to receive a first slice; the first slice is a part of the first semantic model corresponding to the task category and the new modality category;

The second semantic model obtaining module is configured to use the first slice to obtain a second semantic model, so as to perform the pragmatic task on the input information source through the second semantic model.

For the functional units contained in the functional modules in the above edge node, reference may be made to the relevant description of the semantic communication method described above from the edge node side, and this embodiment will not describe it in detail here.

Optionally, this application also proposes a semantic communication device, which can be applied to the cloud of a communication network, and the device can include:

A semantic model update request receiving module, configured to receive a semantic model update request; the semantic model update request includes a new modal category of an input source for requesting an updated semantic model to perform a pragmatic task, and a task of the pragmatic task category;

A first slice obtaining module, configured to obtain a first slice stored corresponding to the task category and the new modality category; the first slice is obtained through a data set belonging to the new modality category , part of the first semantic model trained for the pragmatic task;

The first slice sending module is configured to send the first slice, so that the semantic model update request end can use the first slice to obtain the ability to execute the described input source with the new modality category. A second semantic model for pragmatic tasks.

Regarding the functional units included in the functional modules of the above cloud, reference may be made to the relevant description of the semantic communication method described above from the cloud side, and details are not described in this embodiment here.

It should be noted that the various modules, units, etc. in the above-mentioned apparatus embodiments can all be stored as program modules in the memory of the corresponding side device in the communication network, and executed by the processor in the side device and stored in the memory. The above-mentioned program modules to realize the corresponding functions. For the functions realized by each program module and its combination, as well as the technical effects achieved, you can refer to the description of the corresponding part of the above-mentioned method embodiment executed by the corresponding side device, and this embodiment will not repeat.

The present application also proposes a computer-readable storage medium, on which at least one computer instruction set is stored, and the computer instruction set can be loaded and executed by a processor to implement the above-mentioned semantic communication method executed by the corresponding side of the communication network.

Referring to FIG. 8 , it is a schematic diagram of the hardware structure of an optional example of a computer device suitable for the semantic communication method proposed in this application. The computer device may include a transceiver 81 and a processor 82, wherein:

The transceiver 81 can be used to realize the receiving and sending of information, which can be a communication module supporting a wireless communication network or a wired communication network, such as a WIFI module, 5G/6G (fifth generation mobile communication network/sixth generation mobile communication network) module, wireless radio frequency module, short-distance communication module, GPRS module and other wireless communication modules, or network data lines, etc., can be determined according to the communication mode between different nodes in the communication network. limit.

In the case that the computer devices are configured as devices of different identities in the communication network, the processor 82 can be used to realize the semantic communication method performed by the corresponding side. For example, if the computer device is configured as any edge node in the communication network, it can be used to implement the semantic communication method described above from the edge node side; if the computer device is configured as the cloud of the communication network, it can be used for To implement the semantic communication method described above from the cloud side, the implementation process refers to the description content of the above-mentioned embodiment on the corresponding side, and details are not described in this embodiment.

In the actual application of this application, the above-mentioned processor 82 may be a central processing unit (Central Processing Unit, CPU), a specific application integrated circuit (application-specific integrated circuit, ASIC), a digital signal processor (DSP), an application-specific integrated circuit (ASIC) ), off-the-shelf programmable gate array (FPGA) or other programmable logic devices, etc., the types of processors 82 in different execution objects of the semantic communication method may be the same or different, which is not limited in this application.

Optionally, the above-mentioned semantic communication method may be implemented by program code, and the above-mentioned processor 82 may include a storage device for storing program code for implementing the semantic communication method executed by the corresponding side, and the processor 82 may execute the program code to realize the corresponding Semantic communication method executed on the side.

In some other embodiments, the computer device may also include a memory, which is a device independent of the processor 81, such as a high-speed random access memory, and may also include a non-volatile memory, such as at least one magnetic disk storage device or other volatile memory. A non-volatile solid-state storage device, etc., such as the method described above, can be used to store the program code for implementing the semantic communication method executed by the corresponding side, and the program code is loaded and executed by the processor 82 to realize the semantic communication method executed by the corresponding side.

It should be understood that the structure of the computer device shown in FIG. 8 does not constitute a limitation on the computer device in the embodiment of the present application. In practical applications, the computer device may include more components than those shown in FIG. 8 , or combine certain These components, such as sensor modules composed of various sensors, alarm equipment, power supply modules, etc., when the computer equipment is a terminal equipment, can also include touch sensing units such as touch events on the touch display panel, keyboards, At least one input component such as a mouse, a camera, and a pickup, and at least one output component such as a display, a speaker, a vibration mechanism, and a lamp, etc., can be determined according to the type of computer equipment and its functional requirements, and this application does not list them here.

It should be noted that, in the above-mentioned embodiments, relative terms such as first, second, etc. are only used to distinguish one operation, unit or module from another operation, unit or module, and do not necessarily require Or imply any such actual relationship or order between these units, operations or modules. Furthermore, the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method or system comprising a set of elements includes not only those elements but also other elements not expressly listed. elements, or elements inherent in such a process, method, or system. Without further limitations, an element defined by the phrase "comprising a ..." does not preclude the presence of additional identical elements in the process, method or system comprising said element.

And, as shown in the application and claims, words such as "a", "an", "an" and/or "the" are not specific to the singular and may include the plural, unless the context clearly suggests an exception. Also, unless otherwise specified, "/" means or, for example, A/B can mean A or B; "and/or" in this article is just an association relationship describing associated objects, indicating that there can be three A relationship, for example, A and/or B, can mean: A exists alone, A and B exist simultaneously, and B exists alone.

In addition, the present application illustrates a flow chart of operations performed by the system according to the embodiments of the present application, and the previous or subsequent operations are not necessarily performed in exact order. Instead, various steps may be processed in reverse order or simultaneously. At the same time, other operations can be added to these procedures, or a certain step or steps can be removed from these procedures.

Finally, each embodiment in this specification is described in a progressive or parallel manner, and each embodiment focuses on the difference from other embodiments, and the same and similar parts of each embodiment can be referred to each other. As for the device and computer equipment disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and for the related parts, please refer to the description of the method part.

The above description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the application. Therefore, the present application will not be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (10)

1. A semantic communication method, said method comprising: Obtain new modal categories of input sources for pragmatic tasks; sending a semantic model update request; the semantic model update request includes the task category of the pragmatic task and the new modality category; receiving a first slice; the first slice is a part of a first semantic model corresponding to the task category and the new modality category; Using the first slice, a second semantic model is obtained, so as to perform the pragmatic task on the input information source through the second semantic model. 2. The method according to claim 1, before said obtaining the new modality category of the pragmatic task input source, said method further comprises: Obtain the task category of any pragmatic task to be executed, and the modality category of the input source of the pragmatic task to be processed; Send a semantic model acquisition request; the semantic model acquisition request includes the task category and the modal category; receiving a third semantic model; the third semantic model is a trained semantic model corresponding to the task category and the modality category; Executing the third semantic model to perform the pragmatic task on the input information source to be processed through the third semantic model. 3. The method according to claim 1 or 2, said utilizing said first slice to obtain a second semantic model, comprising: Using the first slice to replace the third slice in the third semantic model to obtain the second semantic model; the third semantic model refers to the implementation of the pragmatic task on the input source of the original modal category semantic model. 4. The method according to claim 1 or 2, in the case of obtaining the new modality category of the pragmatic task input source, the method further comprises: Determine at least one candidate slice that has been stored; the candidate slice is a partial semantic model trained for the pragmatic task; Obtaining a candidate modality category corresponding to the candidate slice; the candidate modality category is a modality category for training a data set corresponding to the semantic model to which the candidate slice belongs; Comparing the new modality category with the candidate modality category to obtain a corresponding comparison result; Determining that the comparison result is that the new modality category is the same as any of the candidate modality categories, obtaining the candidate slice corresponding to the new modality category is determined as the first slice, and performing the using the Describe the first slice, obtain the second semantic model step; If it is determined that the comparison result is that the new modality category is different from all the candidate modality categories, or it is determined that no candidate slice is stored, the step of sending a semantic model update request is performed. 5. A semantic communication method, said method comprising: Receiving a semantic model update request; the semantic model update request is a new modal category including an input source that requests the updated semantic model to perform a pragmatic task, and a task category of the pragmatic task; Obtaining a first slice stored corresponding to the task category and the new modality category; the first slice is trained for the pragmatic task through a data set belonging to the new modality category Part of the first semantic model; The first slice is sent, so that the semantic model update request end uses the first slice to obtain a second semantic model capable of performing the pragmatic task on the input information source with the new modality category. 6. The method of claim 5, further comprising: Receiving a semantic model acquisition request; the semantic model acquisition request includes the task category of the pragmatic task to be executed by the semantic model requested to be acquired, and the modal category of the input source of the pragmatic task to be processed; obtaining a third semantic model corresponding to the task category and the modality category; Send the third semantic model. 7. The method of claim 5 or 6, further comprising: receiving a plurality of semantic models; the plurality of semantic models are for pragmatic tasks of the same task category, obtained through training of data sets of different modality categories; Performing parameter difference analysis on the multiple semantic models to obtain slices for the same pragmatic function included in the multiple semantic models; The respective slices of the plurality of semantic models are associated with the task category and the corresponding modality category and stored. 8. The method according to claim 7, in the case of receiving any one of the semantic model acquisition requests, the method further comprises: Sending the cross-modal slice associated with the task category, so that the semantic model acquisition request end can determine the received cross-modal slice as a candidate slice of the pragmatic task for storage; Wherein, the cross-modal slice refers to a slice contained in semantic models other than the requested semantic model among the plurality of semantic models stored in association with the task category. 9. The method of claim 5 or 6, further comprising: Receiving a semantic model training request; the semantic model training request includes multiple modal categories of different input sources of the pragmatic task to be performed by the semantic model requested to be trained, and the task category of the pragmatic task; obtaining respective datasets of the plurality of modality categories labeled for the task category; According to the data set, train to obtain a semantic model corresponding to the task category and the corresponding modality category; Perform parameter difference analysis on the multiple semantic models obtained through training, and obtain the slices for the same pragmatic function contained in each of the multiple semantic models; The respective slices of the plurality of semantic models are associated with the task category and the corresponding modality category and stored. 10. A computer device comprising a transceiver and a processor, wherein: In the case where the computer device is configured as any edge node in the communication network, the processor is configured to implement: Obtain new modal categories of input sources for pragmatic tasks; sending a semantic model update request; the semantic model update request includes the task category of the pragmatic task and the new modality category; receiving a first slice; the first slice is a part of a first semantic model corresponding to the task category and the new modality category; Obtaining a second semantic model by using the first slice, so as to perform the pragmatic task on the input information source through the second semantic model; Where the computer device is configured as a cloud of the communication network, the processor is configured to: Receiving a semantic model update request; the semantic model update request is a new modal category including an input source that requests the updated semantic model to perform a pragmatic task, and a task category of the pragmatic task; Obtaining a first slice stored corresponding to the task category and the new modality category; the first slice is trained for the pragmatic task through a data set belonging to the new modality category Part of the first semantic model; The first slice is sent, so that the semantic model update request end uses the first slice to obtain a second semantic model capable of performing the pragmatic task on the input information source with the new modality category.