WO2023138231A1

WO2023138231A1 - Residual propagation method and apparatus for network model

Info

Publication number: WO2023138231A1
Application number: PCT/CN2022/136413
Authority: WO
Inventors: 许晓东; 任水迪; 韩书君; 董辰; 王碧舳
Original assignee: 北京邮电大学
Priority date: 2022-01-20
Filing date: 2022-12-04
Publication date: 2023-07-27
Also published as: CN116527561A

Abstract

Provided are a residual propagation method and apparatus for a network model, which relate to the technical field of communications. The specific implementation is: the method is applied to a network, the network comprises a routing path, and the routing path comprises a source node, an intermediate node and a sink node; and the source node stores all model slices in a preset sink node requirement, and the sink node requirement is a requirement of the sink node for the model slices. The residual propagation method comprises: acquiring a routing path (S101); and performing traversal from a sink node to a source node along the routing path, and an intermediate node and/or the source node sending a model slice in a sink node requirement to the sink node (S102).

Description

A residual propagation method and residual propagation device for a network model

technical field

The present disclosure relates to the field of communication technologies, and in particular, to a residual propagation method and a residual propagation device of a network model.

Background technique

In the future intelligent network of all things, network nodes tend to be intelligent. The intelligentization of network nodes has led to rapid expansion of information space, and even dimensional disasters, which has exacerbated the difficulty of representing information carrying space, making it difficult to match traditional network service capabilities with high-dimensional information space. The amount of data transmitted through communication is too large, and the information business service system cannot continue to meet people's needs for complex, diverse, and intelligent information transmission. Using artificial intelligence models to encode, disseminate, and decode business information can significantly reduce the amount of data transmission in communication services and greatly improve the efficiency of information transmission. These models are relatively stable, and have reusability and dissemination. The dissemination and reuse of models will help to enhance network intelligence while reducing overhead and resource waste, forming an intelligent network with extremely intelligent nodes and a minimal network.

Based on the background of extremely intelligent nodes, minimal network, combination of virtual and real, and digital twins in the IDN, what is disseminated in the IDN is no longer just traditional content data, but a relatively stable model generated by calculation. The network has a storage function, and the model is stored in the network, which may be stored on the end user side or in the cloud. Each node can absorb many models on the network to realize self-evolution, which is similar to knowledge distillation. The essence of model propagation is federated learning, which requires support and control from corresponding protocols. Therefore, the technical problem that needs to be solved at present is: how to realize the transmission model in the communication process.

Contents of the invention

The disclosure provides a residual propagation method and a residual propagation device of a network model.

According to a first aspect of the present disclosure, there is provided a method for residual propagation of a network model, which is applied to a network, and the network includes at least one routing path, and all routing paths include a source node and a sink node, and an intermediate node arranged between the source node and the sink node;

Among them, the source node stores all model slices in the preset demand of the sink node, and the demand of the sink node is the demand of the sink node for model slices;

Residual propagation methods include:

Get routing path;

Traversing along the routing path from the sink node to the source node, the intermediate node and/or the source node sends the model slices required by the sink node to the sink node.

According to a second aspect of the present disclosure, there is provided a residual propagation device for a network model, which is applied to a network, and the network includes at least one routing path, and all routing paths include a source node and a sink node, and an intermediate node arranged between the source node and the sink node;

Among them, the source node stores all model slices in the preset sink node requirements, and the sink node requirements are the sink node's demand for model slices;

The residual propagation device includes:

a path obtaining unit, configured to obtain a routing path;

The processing unit is configured to traverse along the routing path from the sink node to the source node, and the intermediate node and/or the source node sends the model slice required by the sink node to the sink node.

According to a third aspect of the present disclosure, an electronic device is provided, including:

at least one processor; and

memory communicatively coupled to at least one processor; wherein,

The memory stores instructions executable by at least one processor, and the instructions are executed by at least one processor, so that the at least one processor can execute the above method.

According to a fourth aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute the above method.

The beneficial effects brought by the technical solution provided by the disclosure are:

The solution provided by the embodiments of the present disclosure performs reverse traversal along the routing path from the sink node to the source node, and for the model slices that need to be transmitted, the traversed intermediate nodes and/or source nodes transmit the necessary parts to the sink nodes according to the needs of the sink nodes, thereby forming model residual propagation, which not only reduces the time delay for the sink nodes to obtain the required model slices, but also reduces data redundancy in the end-to-end communication process, and improves the resource utilization of the overall network.

It should be understood that what is described in this section is not intended to identify key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will be easily understood from the following description.

Description of drawings

The accompanying drawings are used to better understand the present solution, and do not constitute a limitation to the present disclosure. in:

FIG. 1 is a schematic flowchart of a method for residual propagation of a network model according to Embodiment 1 of the present disclosure;

FIG. 2 is a schematic flowchart of step S102 in the method for residual propagation of a network model according to Embodiment 1 of the present disclosure;

FIG. 3 is a routing path diagram without model residual propagation according to Embodiment 1 of the present disclosure;

4-8 are flow charts of model residual propagation in Embodiment 1 of the present disclosure;

FIG. 9 is a routing path diagram after model residual propagation is completed according to Embodiment 1 of the present disclosure;

FIG. 10 is a schematic structural diagram of a residual propagation device provided according to Embodiment 2 of the present disclosure;

FIG. 11 is a block diagram of an electronic device of an embodiment of the present disclosure.

Detailed ways

Exemplary embodiments of the present disclosure are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present disclosure to facilitate understanding, and they should be regarded as exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

In IDN, business information is mainly disseminated through the artificial intelligence model. By using the artificial intelligence model to compress the first business information to be disseminated into the second business information related to the artificial intelligence model, the data traffic in the network is greatly reduced, and the compression efficiency far exceeds the traditional compression algorithm. Wherein, the sending end device uses the preconfigured first model to extract the first service information and obtain the second service information to be transmitted; the sending end device transmits the second service information to the receiving end device. The receiving end device receives the second service information, and uses the pre-configured second model to recover the second service information to obtain the third service information; the third service information restored by the second model will have a slight difference in quality compared with the original first service information, but the two are consistent in content, and the user experience is almost the same. Before the transmitting end device transmits the second service information to the receiving end device, it also includes: an update module judges whether the receiving end device needs to update the second model, and transmits a pre-configured third model to the receiving end device when it is judged that an update is required, and the receiving end device uses the third model to update the second model. Processing business information through pre-trained artificial intelligence models can significantly reduce the amount of data transmission in communication services and greatly improve the efficiency of information transmission. These models are relatively stable, and have reusability and dissemination. Model propagation and reuse will help enhance network intelligence while reducing overhead and resource waste. The model can be divided into several model slices according to different segmentation rules. The above model slices can also be transmitted between different network nodes, and the model slices can be assembled into models. Model slices can be distributed and stored on multiple network nodes. When a network node finds that it lacks or needs to update a certain model or a certain model slice, it can make a request to the surrounding nodes that may have the slice.

Both the transmission of business information and the transmission model occur at the network layer of the communication network, and the communication transmission is performed based on the network layer protocol. The network nodes passing through the path of transmitting service information and transmitting models include the IDR. The functions of the I-Driven Router include but are not limited to business information transmission, model transmission, absorbing model self-update, security protection and other functions. The transmission function of the Intelligent-Driven Router involves the transmission of business information or models from the source node to the sink node, and there are multiple paths between the source node and the sink node. The model transmission function of the Smart-Driven Router can transmit model slices. By rationally arranging model slices to take multiple paths, multiple transmission model slices can be used to improve the model transmission rate.

Embodiment one

Fig. 1 shows a residual propagation method of a network model provided by an embodiment of the present disclosure, wherein, applied to a network, the network includes at least one routing path, and all routing paths include a source node and a sink node, and an intermediate node arranged between the source node and the sink node;

As shown in Figure 1, including:

Step S101, obtaining a routing path;

Step S102, traversing along the routing path from the sink node to the source node, and the intermediate node and/or the source node sends the model slice required by the sink node to the sink node.

The disclosure traverses backwards along the routing path from the sink node to the source node, and for the model slices that need to be transmitted, the traversed intermediate nodes and/or source nodes transmit the necessary parts to the sink nodes according to the needs of the sink nodes, thereby forming model residual propagation, which not only reduces the time delay for the sink nodes to obtain the required model slices, but also reduces the data redundancy in the end-to-end communication process and improves the resource utilization of the overall network.

Specifically, when the routing path is determined, the source node, intermediate node, and sink node on the routing path are also determined at this time. At this time, the transmission nodes can be traversed in reverse order from the sink node to the source node along the routing path. Among them, the transmission node includes the intermediate node and the source node, and only the model slice required by the sink node needs to be transmitted, thus forming the model residual transmission.

Exemplarily, the model slice required by the sink node includes the first model slice and the second model slice. At this time, when the first intermediate node under the sink node has the second model slice and the third model slice, it is only necessary to transmit the second model slice to the sink node. At this time, continue to traverse to the second intermediate node. At this time, only the second model slice exists in the second intermediate node. Since the sink node already has the second model slice, and the second intermediate node is the last intermediate node, skip the second intermediate node and traverse to the source node. The source node has all the model slices required by the sink node. , at this time, the source node sends the first model slice to the sink node.

Specifically, the intermediate node stores model slices, but the model slices in the intermediate nodes may include the existing model slices of the sink node, and the model slices in the intermediate nodes may also include model slices required by the sink node.

For example, the sink node already has the third model slice and the fourth model slice, and at this time the sink node needs to store the first model slice and the second model slice;

At this time, the first intermediate node in the routing path has the second model slice and the third model slice, that is, the first intermediate node at this time has the model slice that the sink node already has - the third model slice, and the model slice that the sink node needs - the second model slice;

At this time, the second intermediate node in the routing path only has the third model slice, that is, the second intermediate node at this time only has the existing model slice of the sink node—the third model slice.

An embodiment of the present disclosure provides a possible implementation, wherein determining the requirements of the sink node includes the following steps:

According to the type and quantity of model slices in the source node and the type and quantity of model slices in the sink node, determine the type and quantity of model slices that the sink node still lacks.

Specifically, since all the model slices required by the sink node are stored in the source node, the model slices that the sink node currently lacks can be found in the source node, and these missing model slices are the sink node's demand for model slices.

Specifically, the model slice is an AI model produced by training each intelligent node in the Intent-Driven Network, which can include various types according to the actual situation and node requirements.

For example, model slices can be edited models and generated animation models, etc., and include but are not limited to classification models, segmentation models, graph neural network models, etc.

The embodiment of the present disclosure provides a possible implementation manner, wherein the source node also stores all model slices existing in the sink node.

Specifically, the source node at this time not only stores all the model slices required by the sink node, but also stores all the model slices in the sink node.

Exemplarily, the information source node at this time includes: 2 classification model slices, 2 segmentation model slices and 3 graph neural network model slices;

At this time, the sink node includes: 1 classification model slice and 1 segmentation model slice;

Therefore, the sink node's demand for model slices at this time is: 1 classification model slice, 1 segmentation model slice, and 3 graph neural network model slices.

Set a unique number for each model slice;

According to the number of the model slice in the source node and the number of the model slice in the sink node, determine the model slice that the sink node still lacks.

For example, set a unique number for each model slice, assuming that there are ten types of model slices, labeled A1-A10;

When the source node not only stores all the model slices required by the sink node, but also stores all the model slices already in the sink node;

The source nodes at this time include: model slice A1, model slice A2, model slice A3, model slice A4, model slice A5, model slice A6, model slice A7, model slice A8, model slice A9 and model slice A10;

At this time, the sink nodes include: model slice A1, model slice A2, model slice A3, model slice A4, model slice A5, model slice A6 and model slice A7;

Therefore, at this time, the sink node's requirements for model slices are: model slice A8, model slice A9, and model slice A10.

The embodiment of the present disclosure provides a possible implementation manner. As shown in FIG. 2, step S102 specifically includes the following steps:

Step S1021, traversing along the routing path from the sink node to the source node;

Step S1022, judging whether the intermediate node has at least one model slice required by the sink node;

If it exists, the intermediate node sends the existing model slices in the demand of the sink node to the sink node, and then executes step S1023;

If it does not exist, continue to traverse along the routing path to the direction of the source node, and return to step S1022 until the requirements of the sink node are met or all intermediate nodes cannot be satisfied after traversing all intermediate nodes.

Step S1023, judging whether the requirements of the sink node are met;

If satisfied, end the traversal;

If it is not satisfied, update the requirements of the sink node, continue to traverse along the routing path to the direction of the source node, and return to step S1022 until the requirements of the sink node are met or all intermediate nodes cannot be satisfied after traversing all intermediate nodes.

Specifically, assuming that a certain node B is a source node, and another node C is a sink node, the two nodes form a certain routing path, and the routing path also includes an intermediate node arranged at least one of the source node B and the source node C. The routing path is defined as a routing path, and there are model slices of different numbers and types at each passing node in the routing path. The transmission node includes the intermediate node and the source node B;

When the sink node C wants to obtain some model slices that it lacks, it does not directly initiate a request to the source node B for transmission, but in reverse order, starting from the previous hop transmission node of the sink node C, and inquires each transmission node in the routing path one by one. The specific steps are as follows:

Determine whether the intermediate node has at least one model slice in the requirements of the sink node:

If the intermediate node does not have the model slice required by the sink node, skip the intermediate node and then go back to the previous hop transmission node to continue searching;

If the intermediate node contains the model slice required by the sink node, the transfer node sends the model slice included in the sink node requirement to the sink node C, and judges whether the model slice in the intermediate node meets the sink node requirement:

If the intermediate node does not meet the requirements of the sink node, update the requirements of the sink node, and then trace back to the previous hop transmission node to find the model slice in the updated sink node requirements;

If the intermediate node meets the requirements of the sink node, that is, the intermediate node at this time contains all the model slices required by the sink node, then the intermediate node sends all the model slices it needs to the sink node C, and terminates the backtracking search process at the same time;

If all the intermediate nodes are searched and the model slice required by the sink node C is still not found, then the source node B will send the model slice in the current sink node demand to the sink node C according to the current sink node demand.

Specifically, the sink node can initiate a request containing the sink node's requirements to the last-hop intermediate node according to its preset sink node requirements, so that the intermediate node can search for the model slice in the sink node's requirements according to the received request:

If the model slice in the requirements of the sink node is found, the found model slice is sent to the sink node, and the requirements of the sink node are updated at the same time, and the request containing the updated requirements of the sink node is forwarded to the previous hop transmission node;

If the model slice required by the sink node is not found, the request is directly forwarded to the previous hop transfer node;

If the requirements of the sink node are not met when the last intermediate node is searched, the last intermediate node will forward the request containing the latest sink node requirements to the source node, and the source node will send the model slice in the latest sink node requirements to the sink node, ending this model residual propagation.

If an intermediate node has all the model slices required in the received request, after it sends the corresponding model slices back to the sink node, it stops updating and forwarding the requests containing the needs, terminates the backtracking search process, and ends this model residual propagation.

Exemplarily, assuming that a certain node B is a source node and another node C is a sink node, the two nodes form a certain routing path, and the routing path also includes an intermediate node arranged at least one of the source node B and the source node C, and the intermediate nodes can be set as D1, D2, and D3, and the routing path is defined as a routing path. There are model slices of different numbers and types at each transmission node passing through the routing path, and the transmission node includes the intermediate node and the source node B;

When the sink node C wants to obtain some model slices that it lacks, it does not directly initiate a request to the source node B for transmission, but in reverse order, starting from the last hop transfer node of the sink node C, and inquiring about each transfer node in the routing path one by one, as shown in Figure 3, the routing path includes: sink node C, intermediate node D1, intermediate node D2, intermediate node D3, and source node B in turn,

Set a unique number for each model slice, assuming there are ten types of model slices, labeled A1-A10;

When the source node B not only includes all the model slices required by the sink node C, but also includes all the existing model slices in the sink node C;

The source node B at this time includes: model slice A1, model slice A2, model slice A3, model slice A4, model slice A5, model slice A6, model slice A7, model slice A8, model slice A9 and model slice A10;

At this time, the sink node C includes: model slice A1, model slice A2, model slice A3, model slice A4, model slice A5, model slice A6 and model slice A7;

And the intermediate node D1 at this time includes: model slice A2, model slice A7 and model slice A8;

At this time, the intermediate node D2 includes: model slice A1, model slice A2 and model slice A9;

The intermediate node D3 at this time includes: model slice A1, model slice A5 and model slice A6;

According to the model slices in the source node B and the sink node C, it can be known that:

At this time, the requirements of the sink node are: model slice A8, model slice A9, and model slice A10. For the convenience of description, the requirement of the sink node at this time is defined as the first requirement.

The specific steps are as follows:

As shown in Figure 4, the sink node C initiates a model slicing request according to the sink node's demand for model slicing, first traverses to the intermediate node D1 under the sink node C, and determines that the intermediate node D1 at this time has the model slice A8 in the first demand, and the intermediate node D1 at this time has the model slice A8 required by the sink node. , for the convenience of illustration, the updated sink node requirement is defined as the second requirement;

As shown in Figure 5, at this time, the request is forwarded directly according to the second requirement to the previous transfer node—the intermediate node D2. It is determined that the intermediate node D2 has the model slice A9 in the second demand, and the intermediate node D2 has the model slice A9 required by the sink node. At this time, as shown in FIG. The sink node requirement is defined as the third requirement;

As shown in Figure 6, at this time, the request is directly forwarded to the previous hop transmission node—the intermediate node D3 according to the third requirement, and it is determined that the intermediate node D3 at this time does not have the model slice A10 in the third requirement;

At this time, all the intermediate nodes have been searched, and the model slice required by the sink node C is still not found, as shown in Figure 7, at this time, the request is directly forwarded to the previous hop transfer node—the source node B, according to the third requirement, as shown in Figure 8, according to the current third requirement, the source node B sends the remaining model slice A10 to the sink node C, and the model residual propagation ends at this time;

As shown in Figure 9, the final sink node C includes model slice A1, model slice A2, model slice A3, model slice A4, model slice A5, model slice A6, model slice A7, model slice A8, model slice A9 and model slice A10.

The embodiment of the present disclosure provides a possible implementation manner, wherein the model slice can be run and stored in all nodes in the network, and can be transmitted between all nodes.

Specifically, the source nodes, sink nodes, and intermediate nodes are all intelligent nodes in the communication system. The intelligent nodes include but are not limited to smartphones, tablet computers, laptops, and edge servers. The above-mentioned intelligent nodes all have strong computing capabilities, can learn and train to generate AI models, can perform hierarchical semantic intelligent information source coding and classification models, and have the ability to absorb many models on the network to achieve self-evolution.

An embodiment of the present disclosure provides a possible implementation manner, wherein the network is a residual network.

It should be noted that the residual network is a convolutional neural network proposed by four scholars from Microsoft Research (Microsoft Research). It won the image classification and object recognition in the 2015 ImageNet Large Scale Visual Recognition Challenge (ILSVRC). The characteristic of the residual network is that it is easy to optimize and can improve the accuracy by adding considerable depth. Its internal residual block uses skip connections, which alleviates the problem of gradient disappearance caused by increasing depth in deep neural networks.

Embodiment two

Fig. 10 shows a residual propagation device of a network model provided by an embodiment of the present disclosure, which is characterized in that it is applied in a network, and the network includes at least one routing path, and all routing paths include a source node and a sink node, and intermediate nodes arranged between the source node and the sink node;

As shown in Figure 10, the residual propagation device includes:

a path obtaining unit 201, configured to obtain a routing path;

The processing unit 201 is configured to traverse along the routing path from the sink node to the source node, and the intermediate node and/or the source node sends the model slices required by the sink node to the sink node.

An embodiment of the present disclosure provides a possible implementation manner, and the processing unit 202 includes:

The demand determination module is configured to determine the type and quantity of model slices that the sink node still lacks according to the type and quantity of model slices in the source node and the type and quantity of model slices in the sink node.

An embodiment of the present disclosure provides a possible implementation manner, wherein the source node further includes all model slices existing in the sink node.

The demand determination module is configured to set a unique number for each model slice, and determine the model slices that the sink node still lacks according to the number of the model slice in the source node and the number of the model slice in the sink node.

A traversal module, configured to traverse along the routing path from the sink node to the source node;

A first judging module, judging whether the intermediate node has at least one model slice required by the sink node;

If it exists, the intermediate node sends the model slice in the existing sink node requirement to the sink node;

If it does not exist, continue to traverse along the routing path to the direction of the source node, so that the intermediate node sends the model slice in the existing sink node demand to the sink node;

The second judging module judges whether the requirements of the sink node are met;

If satisfied, end the traversal;

If it is not satisfied, update the requirements of the sink node and continue to traverse along the routing path to the direction of the source node, so that the intermediate node sends the model slices in the requirements of the sink node to the sink node.

For the embodiments of the present disclosure, the beneficial effects achieved are the same as those of the above embodiment of the method for residual propagation of the network model, and will not be repeated here.

In the technical solution of the present disclosure, the acquisition, storage and application of the user's personal information involved are in compliance with relevant laws and regulations, and do not violate public order and good customs.

According to the embodiments of the present disclosure, the present disclosure also provides an electronic device and a readable storage medium.

The electronic equipment, including:

at least one processor; and

memory communicatively coupled to at least one processor; wherein,

The electronic device performs reverse traversal along the routing path from the sink node to the source node, and for the model slices that need to be transmitted, the traversed intermediate nodes and/or source nodes transmit the existing necessary parts to the sink node according to the needs of the sink node, thereby forming model residual propagation, which not only reduces the time delay for the sink node to obtain the required model slices, but also reduces data redundancy in the end-to-end communication process and improves the resource utilization of the overall network.

The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to make a computer execute the method provided by the embodiments of the present disclosure.

The readable storage medium performs reverse traversal along the routing path from the sink node to the source node, and for the model slices that need to be transmitted, the traversed intermediate nodes and/or source nodes transmit the existing necessary parts to the sink node according to the needs of the sink node, thereby forming model residual propagation, which not only reduces the time delay for the sink node to obtain the required model slices, but also reduces data redundancy in the end-to-end communication process, and improves the resource utilization of the overall network.

FIG. 11 shows a schematic block diagram of an example electronic device 300 that may be used to implement embodiments of the present disclosure. Electronic device is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are by way of example only, and are not intended to limit implementations of the disclosure described and/or claimed herein.

As shown in FIG. 11 , the device 300 includes a computing unit 301, which can execute various appropriate actions and processes according to a computer program stored in a read-only memory (ROM, Read-Only Memory) 302 or a computer program loaded from a storage unit 308 into a random access memory (RAM, Random Access Memory) 303. In the RAM 303, various programs and data necessary for the operation of the device 300 can also be stored. The computing unit 301, ROM 302, and RAM 303 are connected to each other through a bus 304. An input/output (I/O, Input/Output) interface 307 is also connected to the bus 304 .

Multiple components in the device 300 are connected to the I/O interface 305, including: an input unit 306, such as a keyboard, a mouse, etc.; an output unit 307, such as various types of displays, speakers, etc.; a storage unit 308, such as a magnetic disk, an optical disk, etc.; and a communication unit 309, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 309 allows the device 300 to exchange information/data with other devices over a computer network such as the Internet and/or various telecommunication networks.

The computing unit 301 may be various general-purpose and/or special-purpose processing components having processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, a central processing unit (CPU, Central Processing Unit), a graphics processing unit (GPU, Graphics Processing Unit), various dedicated artificial intelligence (AI, Artificial Intelligence) computing chips, various computing units that run machine learning model algorithms, digital signal processors (DSP, Digital Signal Processing), and any appropriate processors, controllers, microcontrollers, etc. The calculation unit 301 executes various methods and processes described above, such as the residual propagation method of the network model. For example, in some embodiments, the residual propagation method for network models may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as storage unit 307 . In some embodiments, part or all of the computer program may be loaded and/or installed on the device 300 via the ROM 302 and/or the communication unit 309. When the computer program is loaded into the RAM 303 and executed by the computing unit 301, one or more steps of the residual propagation method for the network model described above can be performed. Alternatively, in other embodiments, the computing unit 301 may be configured in any other appropriate way (for example, by means of firmware) to execute the residual propagation method of the network model.

Various implementations of the systems and technologies described above in this paper can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGA, Field Programmable Gate Array), application specific integrated circuits (ASIC, Application Specific Integrated Circuit), application specific standard products (ASSP, Application Specific Standard Product), systems on chips (SOC, System on Chip), load programmable logic devices (CPLD, Comp lex Programmable Logic Device), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include being implemented in one or more computer programs executable and/or interpreted on a programmable system comprising at least one programmable processor, which may be a special purpose or general purpose programmable processor, capable of receiving data and instructions from and transmitting data and instructions to a storage system, at least one input device, and at least one output device.

Program codes for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to processors or controllers of general-purpose computers, special purpose computers, or other programmable data processing devices, so that the program codes cause the functions/operations specified in the flowcharts and/or block diagrams to be implemented when executed by the processors or controllers. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media would include one or more wire-based electrical connections, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), EPROM (Electrical Programmable Read-Only Memory or flash memory), optical fiber, Compact Disc Read-Only Memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing. combination.

To provide for interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (Cathode Ray Tube, cathode ray tube) or LCD (Liquid Crystal Display, liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic input, voice input, or tactile input.

The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication, eg, a communication network. Examples of communication networks include: Local Area Network (LAN, Local Area Network), Wide Area Network (WAN, Wide Area Network) and the Internet.

A computer system may include clients and servers. Clients and servers are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, a server of a distributed system, or a server combined with a blockchain.

It should be understood that steps may be reordered, added or deleted using the various forms of flow shown above. For example, each step described in the present disclosure may be executed in parallel, sequentially, or in a different order, as long as the desired result of the technical solution disclosed in the present disclosure can be achieved, no limitation is imposed herein.

The specific implementation manners described above do not limit the protection scope of the present disclosure. It should be apparent to those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made depending on design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present disclosure shall be included within the protection scope of the present disclosure.

Claims

A method for residual propagation of a network model, characterized in that it is applied to a network, and the network includes at least one routing path, and all of the routing paths include a source node and a sink node, and an intermediate node arranged between the source node and the sink node;

Wherein, the source node stores all model slices in the preset sink node requirements, and the sink node requirements are the sink node's requirements for model slices;

The residual propagation method includes:

Obtain the routing path (S101);

Traversing along the routing path from the sink node to the source node, the intermediate node and/or the source node sends the model slices required by the sink node to the sink node (S102).
The residual propagation method according to claim 1, wherein determining the demand of the sink node comprises the following steps:

Determine the type and number of model slices that the sink node still lacks according to the type and number of model slices in the source node and the type and number of model slices in the sink node.
The residual propagation method according to claim 1, wherein the source node also stores all existing model slices in the sink node.
The residual propagation method according to claim 1, wherein determining the demand of the sink node comprises the following steps:

Set a unique number for each model slice;

Determine the model slice that the sink node still lacks according to the number of the model slice in the source node and the number of the model slice in the sink node.
The method for residual propagation according to any one of claims 1-4, wherein the traversing along the routing path from the sink node to the source node, the intermediate node and/or the source node sending the model slices required by the sink node to the sink node includes:

Traverse along the routing path from the sink node to the source node;

Judging whether the intermediate node has at least one model slice required by the sink node;

If it exists, the intermediate node sends the existing model slices required by the sink node to the sink node;

If it does not exist, continue to traverse along the routing path to the direction of the source node, so that the intermediate node sends the existing model slice in the sink node requirement to the sink node;

judging whether the requirements of the sink node are met;

If satisfied, end the traversal;

If it is not satisfied, update the requirements of the sink node, and continue to traverse along the routing path to the direction of the source node, so that the intermediate node sends the model slices in the requirements of the sink node to the sink node.
The residual propagation method according to claim 1, wherein the model slice can be run and stored in all nodes in the network, and can be transmitted between all nodes.
A residual propagation device for a network model, characterized in that it is applied in a network, and the network includes at least one routing path, and all the routing paths include a source node and a sink node, and an intermediate node arranged between the source node and the sink node;

Wherein, the source node stores all model slices in the preset sink node requirements, and the sink node requirements are the sink node's requirements for model slices;

The residual propagation device includes:

a path obtaining unit (201), configured to obtain the routing path;

A processing unit (202), configured to traverse along the routing path from the sink node to the source node, and the intermediate node and/or the source node send the model slices required by the sink node to the sink node.
The residual propagation device according to claim 7, wherein the processing unit (202) comprises:

A requirement determination module, configured to determine the type and quantity of model slices that the sink node still lacks according to the type and quantity of model slices in the source node and the type and quantity of model slices in the sink node.
The residual propagation device according to claim 7, wherein the processing unit (202) comprises:

The demand determination module is configured to set a unique number for each model slice, and determine the model slices that the sink node still lacks according to the number of model slices in the source node and the number of model slices in the sink node.
The residual propagation device according to any one of claims 7-9, wherein the processing unit (202) comprises:

A traversal module, configured to traverse along the routing path from the sink node to the source node;

A first judging module, judging whether the intermediate node has at least one model slice required by the sink node;

If it exists, the intermediate node sends the existing model slices required by the sink node to the sink node;

If it does not exist, continue to traverse along the routing path to the direction of the source node, so that the intermediate node sends the existing model slice in the sink node requirement to the sink node;

A second judging module, judging whether the requirements of the sink node are met;

If satisfied, end the traversal;

If it is not satisfied, update the requirements of the sink node, and continue to traverse along the routing path to the direction of the source node, so that the intermediate node sends the model slices in the requirements of the sink node to the sink node.
An electronic device comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

The memory stores instructions executable by the at least one processor, the instructions are executed by the at least one processor, so that the at least one processor can perform the method of any one of claims 1-6.
A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to perform the method according to any one of claims 1-6.