[go: up one dir, main page]

CN113692052A - Network edge machine learning training method - Google Patents

Network edge machine learning training method Download PDF

Info

Publication number
CN113692052A
CN113692052A CN202110897051.9A CN202110897051A CN113692052A CN 113692052 A CN113692052 A CN 113692052A CN 202110897051 A CN202110897051 A CN 202110897051A CN 113692052 A CN113692052 A CN 113692052A
Authority
CN
China
Prior art keywords
channel
machine learning
data
task
edge server
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110897051.9A
Other languages
Chinese (zh)
Inventor
郭棉
李佳锐
单纯
柳秀山
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong Polytechnic Normal University
Original Assignee
Guangdong Polytechnic Normal University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guangdong Polytechnic Normal University filed Critical Guangdong Polytechnic Normal University
Priority to CN202110897051.9A priority Critical patent/CN113692052A/en
Publication of CN113692052A publication Critical patent/CN113692052A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W72/00—Local resource management
    • H04W72/50—Allocation or scheduling criteria for wireless resources
    • H04W72/53—Allocation or scheduling criteria for wireless resources based on regulatory allocation policies
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/12—Protocols specially adapted for proprietary or special-purpose networking environments, e.g. medical networks, sensor networks, networks in vehicles or remote metering networks
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W72/00—Local resource management
    • H04W72/50—Allocation or scheduling criteria for wireless resources
    • H04W72/54—Allocation or scheduling criteria for wireless resources based on quality criteria
    • H04W72/542—Allocation or scheduling criteria for wireless resources based on quality criteria using measured or perceived quality

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Medical Informatics (AREA)
  • Quality & Reliability (AREA)
  • Theoretical Computer Science (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本发明公开了一种网络边缘机器学习训练方法，包括：边缘服务器接收机器学习训练任务，读取任务信息；边缘服务器根据任务信息确定数据源集合；边缘服务器向所确定的数据源集合所对应的物联网终端发送数据采集请求；物联网终端接收数据采集请求，并向基站发送信道分配请求；基站接收信道分配请求，为所述物联网终端分配信道；物联网终端通过所分配的信道将本地数据集发送给边缘服务器；边缘服务器接收本地数据集，根据本地数据集执行机器学习训练任务，返回训练结果。本发明实现将机器学习训练从云计算中心下沉到网络边缘，有效减少数据传输延迟，提高数据隐私性及安全性。

Figure 202110897051

The invention discloses a network edge machine learning training method, comprising: an edge server receives a machine learning training task and reads task information; the edge server determines a data source set according to the task information; The IoT terminal sends a data collection request; the IoT terminal receives the data collection request and sends a channel allocation request to the base station; the base station receives the channel allocation request and allocates a channel for the IoT terminal; the IoT terminal transfers the local data through the allocated channel. The data set is sent to the edge server; the edge server receives the local data set, performs the machine learning training task according to the local data set, and returns the training result. The invention realizes that the machine learning training is sinking from the cloud computing center to the network edge, effectively reducing the delay of data transmission, and improving the privacy and security of the data.

Figure 202110897051

Description

Network edge machine learning training method
Technical Field
The invention relates to the field of edge learning, in particular to a network edge machine learning training method.
Background
New businesses of Internet of things such as Internet of vehicles, telemedicine, virtual reality and the like often need model training through machine learning. At present, the training process of machine learning is mainly executed in a cloud computing center. However, the cloud computing paradigm has been difficult to meet the ever-increasing business demands. The reason is that the cloud computing center is often far away from a data source required by machine learning model training, and mass data are transmitted to a remote cloud computing data center through a network, so that the data are subjected to long network delay and cannot meet the low-delay requirement of a service, and the privacy and the safety of the data are difficult to guarantee.
With the advancement of 5G business, 5G + edge computing has been considered as an effective paradigm to deal with "massive data, ultra-low latency, data privacy and security". The computing server is deployed at the edge of the network, data generated by the Internet of things terminal and the mobile terminal are transmitted to the edge server of the same edge network, and the edge server executes machine learning training, so that not only can network transmission delay be reduced and user experience of Internet of things services be improved, but also data do not leave an edge network area where a data source is located, and privacy and safety of the data are greatly improved.
Generally, data of distributed data sources under the same edge network needs to be aggregated to an edge server before machine learning training is performed. The faster the data is aggregated, the earlier the machine learning training can be started, whereas the larger the data aggregation delay, the larger the waiting delay of the machine learning training is, and the worse the user experience is. However, due to the limited radio resources and the dynamic nature of the radio environment, uploading data of multiple internet of things terminals to the edge server at the same time poses many challenges. In the existing method adopting orthogonal multiple access, because the number of channels is far lower than that of terminals of the Internet of things, the terminals of the Internet of things need to queue for competing for wireless resources, and larger network transmission waiting delay is caused; in addition, according to a non-orthogonal multiple access method provided by researchers, the intra-channel interference between terminals of the internet of things is large, the transmission power of the terminals of the internet of things, the signal decoding sequence of a base station side and the like directly affect the data transmission rate of each terminal of the internet of things, and in the method, the data uploading delay difference between the terminals of the internet of things is large. In fact, the data aggregation delay is affected most by the maximum upload delay among the data upload delays of the terminals of the distributed internet of things, and reducing the maximum delay is the key for reducing the data aggregation delay.
In summary, how to combine the multiple access technology of the base station to quickly converge distributed data to the edge server, how to cooperate with the resources of the end (the terminal of the internet of things), the edge (the edge server) and the base station to support the realization of machine learning training at the network edge is a key problem to be urgently solved for promoting the application of new services of the internet of things such as car networking, telemedicine and the like.
Disclosure of Invention
The present invention is directed to overcome at least one of the above-mentioned drawbacks (disadvantages) of the prior art, and to provide a network edge machine learning training method, which is used to solve the problems of long network transmission delay, and low data privacy and security of machine learning training data.
The invention adopts the technical scheme that a network edge machine learning training method comprises the following steps:
the method comprises the steps that an edge server receives a machine learning training task and reads task information of the machine learning training task;
the edge server determines a data source set according to the task information, wherein one data source in the data source set corresponds to one Internet of things terminal in an edge network where the edge server is located;
the edge server sends a data acquisition request to the Internet of things terminal corresponding to the determined data source set;
the Internet of things terminal receives the data acquisition request and sends a channel allocation request to a base station of the edge network according to the data acquisition request;
the base station receives the channel allocation request and allocates channels to the Internet of things terminal according to the channel allocation request;
the terminal of the Internet of things sends a local data set to the edge server through the distributed channel;
and the edge server receives the local data set, executes a machine learning training task according to the local data set and returns a training result.
In the invention, an edge server is deployed at the edge of a network, a local data set generated by an Internet of things terminal or other mobile terminals is transmitted to the edge server of the same edge network through wireless channel resources distributed by a base station, and the edge server receives and executes a machine learning training task from a user according to the local data set, so that the machine learning training task sinks from a cloud computing center to the edge of the network, thereby not only reducing network transmission delay and improving the user experience of the Internet of things service, but also ensuring that the data does not leave the edge network area where the data source is located, and greatly improving the privacy and the safety of the data.
Further, the task information comprises a data type required by the task; the edge server determines a data source set according to the task information, and the method comprises the following steps:
the edge server counts all the Internet of things terminals of an edge network where the edge server is located and the data types collected by all the Internet of things terminals;
and the edge server selects the Internet of things terminal with the same collected data type as the data type required by the task, and determines the data source set according to the selected Internet of things terminal.
According to the method for determining the terminal of the Internet of things according to the data type required by the task in the task information, the edge server can be effectively prevented from carrying out communication interaction with the terminal of the Internet of things without corresponding type data, and waste of network bandwidth resources and energy consumption of terminal equipment of the Internet of things is avoided.
Further, the sending, by the edge server, a data acquisition request to the internet of things terminal corresponding to the determined data source set includes:
the edge server generates a machine learning training task identification number according to the received machine learning training task;
the edge server generates a data acquisition request comprising the machine learning training task identification number according to the machine learning training task identification number, and sends the data acquisition request to the Internet of things terminal corresponding to the confirmed data source set;
the internet of things terminal receives the data acquisition request and sends a channel allocation request to a base station of the edge network according to the data acquisition request, and the method comprises the following steps:
the internet of things terminal receives the data acquisition request, generates a channel allocation request comprising the machine learning training task identification number according to the data acquisition request, and sends the channel allocation request to a base station of the edge network;
the base station receives the channel allocation request, allocates channels for the internet of things terminal according to the channel allocation request, and the method comprises the following steps:
the base station receives the channel allocation request, and the terminals of the Internet of things with the same machine learning training task identification number in the received channel allocation request form a terminal set;
the base station selects a corresponding channel set according to the machine learning training task identification number in the channel allocation request, wherein different machine learning training task identification numbers correspond to different channel sets;
and the base station distributes the selected channels in the channel set for each Internet of things terminal in the terminal set.
In the invention, before the base station distributes the channels to the terminals of the Internet of things, the terminals of the Internet of things with the same machine learning training task identification number in the received channel distribution request form a terminal set, and channel resources are distributed to the terminals of the Internet of things with the same machine learning training task identification number.
Further, the allocating, by the base station, the selected channel in the channel set for each internet of things terminal in the terminal set includes:
the base station performs initialization channel allocation on each Internet of things terminal in the terminal set according to the selected channel in the channel set to obtain an initialization channel allocation result;
the base station calculates data uploading delay of each Internet of things terminal in the terminal set on different distribution channels according to the initialized channel distribution result, wherein the data uploading delay of each Internet of things terminal on different distribution channels is the delay of each Internet of things terminal for sending a local data set to the edge server through different distribution channels;
the base station updates the initialized channel allocation result according to the data uploading delay to obtain an updated channel allocation result;
and the base station distributes the selected channels in the channel set for each Internet of things terminal in the terminal set according to the updated channel distribution result.
Further, the base station performs initialization channel allocation on each internet of things terminal in the terminal set according to the selected channel in the channel set to obtain an initialization channel allocation result, including:
the base station calculates channel power gains of different channels corresponding to each Internet of things terminal in the terminal set, wherein the channel power gains corresponding to the different channels are power gains of the Internet of things terminals for sending local data sets to the edge server through the different channels in the channel set;
and the base station allocates the channel corresponding to the maximum power gain of the channel of each Internet of things terminal to the Internet of things terminal to obtain an initialized channel allocation result.
In the invention, through the initial channel allocation of the base station, each Internet of things terminal requesting channel allocation is firstly allocated to a channel with the best channel quality, namely, the channel with the largest channel power gain of the Internet of things terminal.
Further, the base station calculates data upload delay of each internet of things terminal in the terminal set on different allocation channels according to the initialization channel allocation result, and the method includes:
the base station adopts a formula Dm＝Lm/Rm，nCalculating data uploading delay of each Internet of things terminal in the terminal set, wherein LmThe size R of a local data set of any one Internet of things terminal m in the channel allocation Internet of things terminal set is representedm，nRepresenting the data uploading rate of the terminal m of the Internet of things in the allocated channel n;
in the above formula, Rm，nCan be represented by formula
Figure RE-GDA0003204350280000041
Determining that B represents the bandwidth of the allocation channel n, Pm，nRepresenting the transmitting power h of the terminal m of the Internet of things in the channel nm，nRepresenting the channel power gain of the terminal m of the Internet of things on the channel n, the size Lm of a local data set of the terminal m of the Internet of things and the transmitting power P of the terminal m of the Internet of things on the channel nm，nMay be included in the channel allocation request;
Figure RE-GDA0003204350280000042
representing the inter-channel interference of the terminal m of the internet of things on the channel n,
Figure RE-GDA0003204350280000043
can be represented by formula
Figure RE-GDA0003204350280000044
Determination of γm，nRepresenting the intra-channel interference, gamma, of the terminal m of the Internet of things in the channel nm，nThen again can pass through the formula
Figure RE-GDA0003204350280000045
Determination of N0Representing the Gaussian white noise interference of said channel, wherein N represents the assigned channel set, function
Figure RE-GDA0003204350280000051
Representing a binary indicator function, when the variable X takes the value true,
Figure RE-GDA0003204350280000052
if not, then,
Figure RE-GDA0003204350280000053
collection
Figure RE-GDA0003204350280000054
Representing the set of terminals of the Internet of things distributed to the channel n ', k representing any one terminal of the Internet of things distributed to the channel n', hk，n′Representing the channel power gain, P, of the terminal k of the Internet of things in a channel nk，nRepresenting the transmitting power of the terminal k of the Internet of things in the channel n', and collecting
Figure RE-GDA0003204350280000055
Representing the set of terminals of the Internet of things distributed to the channel n, j representing any terminal of the Internet of things which is distributed to the channel n and is not equal to the terminal m of the Internet of things, hj，nRepresenting the channel power gain, P, of the terminal j of the Internet of things in the channel nj，nAnd representing the transmitting power of the terminal j of the Internet of things in the channel n.
Further, the updating, by the base station, the initialized channel allocation result according to the data upload delay to obtain an updated channel allocation result includes:
the base station selects the Internet of things terminal m' with the largest data uploading delay in the terminal set;
the base station searches a new channel n 'in the selected channel set for the terminal m' of the Internet of things;
the base station judges whether the new channel n' simultaneously satisfies a condition a and a condition b:
condition a: the new channel n ' is a channel with the smallest data uploading delay of the internet of things terminal m ' in the selected channel set except for an initialized distribution channel n ', and the initialized distribution channel n ' is a channel distributed by the internet of things terminal m ' in the initialized channel distribution result;
condition b: the data uploading delay of the internet of things terminal m 'on the new channel n' is lower than that of the internet of things terminal m 'on the initialized distribution channel n';
if so, updating the initialized distribution channel n 'to a new channel n', continuing to calculate the data uploading delay of each Internet of things terminal in the terminal set on different distribution channels, and if not, taking the initialized channel distribution result as the updated channel distribution result.
In the invention, the base station continuously adjusts and optimizes the channel allocation of the Internet of things terminal with the maximum data uploading delay, so that the maximum data uploading delay gradient of the Internet of things terminal is reduced and finally reaches an optimized value, thereby reducing the data acquisition delay of a data set required by machine learning training, enabling a machine learning training task to be started on an edge server quickly and improving the user experience of the Internet of things service corresponding to the machine learning training.
Further, the task information also comprises a task type, a machine learning algorithm adopted by the task and a task precision requirement;
the edge server receives the local data set, executes a machine learning training task according to the local data set, and returns a training result, wherein the training result comprises:
the edge server receiving the local data set;
the edge server converges the local data set;
the edge server takes the converged local data set as a data source and executes machine learning training according to the task type and a machine learning algorithm adopted by the task;
and the edge server judges whether the precision of the machine learning training is not lower than the precision requirement of the task, if so, stores the training result, ends the training, otherwise, continues to execute the step of performing the machine learning training by taking the gathered local data set as a data source according to the type of the task and the machine learning algorithm adopted by the task.
Further, the task information also comprises the maximum delay tolerable by the task;
after the edge server receives the machine learning training task and reads the task information of the machine learning training task, the method further comprises the following steps: the edge server starts a machine learning training task timer according to the maximum delay tolerable by the task, and the machine learning training task timer takes the maximum delay tolerable by the task as a threshold value;
after the edge server uses the aggregated local data set as a data source, and executes machine learning training according to the task type and a machine learning algorithm adopted by the task, and before judging whether the precision of model training is not lower than the precision requirement of the model training, the method further comprises the following steps:
and judging whether the machine learning training task timer is overtime, if so, storing a training result, ending the training, and if not, continuously judging whether the precision of the model training is not lower than the precision requirement of the model training.
Further, after the edge server sends a data acquisition request to the internet of things terminal corresponding to the confirmed data source set, the method further includes: the edge server starts a data acquisition timer, and the threshold value of the data acquisition timer is smaller than that of the machine learning training task timer;
after the edge server receives the local data set and before the edge server aggregates the local data set, the method includes:
the edge server judges whether local data sets of all corresponding Internet of things terminals in a data source set are received or not, if yes, the edge server uses the gathered local data sets as data sources and executes machine learning training according to the task type and a machine learning algorithm adopted by the task, if not, the edge server continuously judges whether the data acquisition timer is overtime or not, if yes, the edge server uses the gathered local data sets as data sources and executes machine learning training according to the task type and the machine learning algorithm adopted by the task, and if not, the edge server continuously executes the edge server to receive the local data sets.
In the invention, the machine learning training task timer and the threshold thereof can effectively control the delay of the whole training task to meet the low-delay requirement of the service of the Internet of things, and the data acquisition timer and the threshold thereof can effectively master the maximum delay of the data acquisition stage and reserve sufficient time for the execution of the subsequent machine learning training task process.
1. According to the invention, the terminal data of the Internet of things is transmitted to the edge server through the channel distributed by the base station, and machine learning training is directly carried out on the edge server, so that a machine learning training task sinks to the edge of the network from the cloud computing center, thus effectively reducing network transmission delay, improving the user experience of the business of the Internet of things, enabling the data not to leave the edge network area where the data source is located, and greatly improving the privacy and the safety of the data;
2. according to the invention, the base station updates and optimizes the channel allocation result of the terminal of the Internet of things through the data uploading delay, so that the data uploading delay and the data acquisition delay are effectively reduced, the machine learning training can be quickly started at the edge server, and the user experience is improved;
3. the invention sets the machine learning training task timer and the data acquisition timer, effectively controls the data acquisition delay and the total delay of the machine learning training task, and can meet the low-delay service requirement of the Internet of things service.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
FIG. 2 is a flowchart illustrating the method of step S107 according to the present invention.
FIG. 3 is a flowchart illustrating a step S105 of the method of the present invention.
FIG. 4 is a flowchart illustrating a step S203 of the method of the present invention.
FIG. 5 is a flowchart illustrating the method of step S303 according to the present invention.
Fig. 6 is a schematic diagram of an edge network model applicable to the present invention.
Detailed Description
The drawings are only for purposes of illustration and are not to be construed as limiting the invention. For a better understanding of the following embodiments, certain features of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product; it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The embodiments of the present invention are all applied to a multiple-input multiple-output-non-orthogonal multiple access (MIMO-NOMA) edge network, where the edge network includes a plurality of internet of things terminals, an edge server and a base station, as shown in fig. 6, the internet of things terminals are randomly distributed in the coverage area of the edge network, the edge server is located at the core of the edge network, that is, near the base station, a high-speed communication bandwidth is provided between the edge server and the base station, and the internet of things terminals and the edge server communicate with each other through the base station. The method comprises the following steps that an Internet of things terminal acquires specific data (such as temperature, environmental data, industrial data, automatic driving data and other industry application data) near the position of the Internet of things terminal, and the data is taken as a data source; the edge server has computational processing power and is loaded with a program for machine learning training. In order to sink machine learning training from a cloud computing center to the network edge, reduce network transmission delay and avoid privacy and safety hidden dangers of data transmission in a public network, data collected by an internet of things terminal are gathered to an edge server, and machine learning training is executed on the edge server side.
Example 1
As shown in fig. 1, the present embodiment provides a network edge machine learning training method, including the following steps:
s101: the edge server receives a machine learning training task, reads task information of the machine learning training task, wherein the task information comprises a task type T, a machine learning algorithm (such as a convolutional neural network, a long-short memory neural network and the like) adopted by the task, and a data type required by the task
Figure RE-GDA0003204350280000081
(for example,
Figure RE-GDA0003204350280000082
or other sensor data, etc.), task accuracy requirement a, maximum delay D tolerable for the taskMaxThe maximum delay D tolerable by the edge server according to the taskMaxStarting a machine learning training task timer, wherein the machine learning training task timer takes the maximum delay tolerable by the task as a threshold value;
s102: the edge server determines a data source set according to the task information, wherein one data source in the data source set corresponds to one Internet of things terminal in an edge network where the edge server is located;
further, the step S102 includes:
the edge server counts all internet of things terminals of an edge network where the edge server is located
Figure RE-GDA0003204350280000083
And the type of data collected by each Internet of things terminal
Figure RE-GDA0003204350280000084
The edge server selects the internet of things terminal with the same collection data type as the data type required by the task, and determines the data source set according to the selected internet of things terminal, namely, for the internet of things terminal
Figure RE-GDA0003204350280000085
The corresponding terminal m of the Internet of things if the data type of the terminal m meets the requirement
Figure RE-GDA0003204350280000086
It is put into the data source set
Figure RE-GDA0003204350280000087
S103: the edge server sends a data acquisition request to the Internet of things terminal corresponding to the determined data source set, the edge server starts a data acquisition timer, and the threshold value of the data acquisition timer is smaller than that of the machine learning training task timer;
s104: the Internet of things terminal receives the data acquisition request and sends a channel allocation request to a base station of the edge network according to the data acquisition request;
s105: the base station receives the channel allocation request, and allocates channels to the terminals of the Internet of things according to the channel allocation request
Figure RE-GDA0003204350280000088
Wherein the vector
Figure RE-GDA0003204350280000089
Indicates the channel allocation result, N1、N2、...、NMRepresenting channels allocated to the internet of things terminals 1, 2,. and M, wherein 1, 2,. and M correspond to the internet of things terminals in the data source set M;
s106: the terminal of the Internet of things sends a local data set to the edge server through the distributed channel;
s107: and the edge server receives the local data set, executes a machine learning training task according to the local data set and returns a training result.
Further, as shown in fig. 2, the step S107 includes:
s171: the edge server receiving the local data set;
s172: the edge server judges whether local data sets of all corresponding internet of things terminals in the data source set are received or not, if yes, step S174 is executed, and if not, step S173 is executed;
s173: judging whether the data acquisition timer is overtime, if so, executing step S174, otherwise, returning to step S171;
s174: the edge server aggregates the local DATA sets, i.e., let DATA ═ { X, Y } represent the DATA fields of the machine learning training, DATAm＝{Xm，YmRepresents a terminal from the Internet of things
Figure RE-GDA0003204350280000091
Local DATA set of, DATAmAfter DATA cleaning and DATA alignment, the DATA is put into a DATA field DATA, so that
Figure RE-GDA0003204350280000092
S175: the edge server takes the converged local DATA set DATA as a DATA source and executes machine learning training according to the task type T and a machine learning algorithm adopted by the task;
s176: judging whether the machine learning training task timer is overtime, namely, ordering DtotRepresenting a time interval from the receipt of the machine learning training task from the edge server to the current time, and determining Dtot≥DMaxIf yes, go to step S178, otherwise go to step S177;
s177: the edge server judges whether the precision of the machine learning training is not lower than the task precision requirement, if so, the step S178 is executed, and if not, the step S175 is executed;
s178: and storing the training result and finishing the training.
Example 2
In the embodiment, on the basis of embodiment 1, a plurality of sub-steps are further expanded.
Further, the step S103 of sending, by the edge server, the data acquisition request to the internet of things terminal corresponding to the confirmed data source set includes:
the edge server generates a machine learning training task identification number ID according to the received machine learning training taskXX；
The edge server identifies the ID according to the machine learning training taskXXGenerating an ID including the machine learning training task identifierXXAnd sending the data acquisition request to the internet of things terminal corresponding to the confirmed data source set.
Further, the step S104 includes:
the Internet of things terminal receives the data acquisition request and generates the machine learning training task identification number ID according to the data acquisition requestXXSending the channel allocation request to a base station of the edge network;
further, as shown in fig. 3, the step S105 includes:
s201: the base station receives the channel allocation request, and the terminals of the Internet of things with the same machine learning training task identification number in the received channel allocation request form a terminal set;
s202: the base station selects a corresponding channel set according to the machine learning training task identification number in the channel allocation request, wherein different machine learning training task identification numbers correspond to different channel sets;
s203: and the base station distributes the selected channels in the channel set for each Internet of things terminal in the terminal set.
Example 3
In this embodiment, on the basis of embodiments 1 and 2, a plurality of sub-steps are further expanded.
Further, as shown in fig. 4, the step S203 includes:
s301: the base station performs initialization channel allocation on each Internet of things terminal in the terminal set according to the selected channel in the channel set to obtain an initialization channel allocation result;
specifically, the step S301 executes the following process:
the base station calculates channel power gains of different channels corresponding to each Internet of things terminal in the terminal set, wherein the channel power gains corresponding to the different channels are power gains of the Internet of things terminals for sending local data sets to the edge server through the different channels in the channel set;
the base station allocates the channel corresponding to the maximum power gain of the channel of each Internet of things terminal to the Internet of things terminal to obtain an initialized channel allocation result;
s302: the base station calculates data uploading delay of each Internet of things terminal in the terminal set on different distribution channels according to the initialized channel distribution result, wherein the data uploading delay of each Internet of things terminal on different distribution channels is the delay of each Internet of things terminal for sending a local data set to the edge server through different distribution channels;
specifically, the base station adopts a formula Dm＝Lm/Rm，nCalculating data uploading delay of each Internet of things terminal in the terminal set, wherein LmThe size R of a local data set of any one Internet of things terminal m in the channel allocation Internet of things terminal set is representedm，nRepresenting the data uploading rate of the terminal m of the Internet of things in the allocated channel n;
in the above formula, Rm，nCan be represented by formula
Figure RE-GDA0003204350280000101
Determining that B represents the bandwidth of the allocation channel n, Pm，nRepresenting the transmitting power h of the terminal m of the Internet of things in the channel nm，nRepresenting the channel power gain of the terminal m of the Internet of things in the channel n and the size L of the local data set of the terminal m of the Internet of thingsmAnd the transmitting power P of the terminal m of the Internet of things on the channel nm，nCan be included inThe channel allocation request;
Figure RE-GDA0003204350280000111
representing the inter-channel interference of the terminal m of the internet of things on the channel n,
Figure RE-GDA0003204350280000112
can be represented by formula
Figure RE-GDA0003204350280000113
Determination of γm，nRepresenting the intra-channel interference, gamma, of the terminal m of the Internet of things in the channel nm，nThen again can pass through the formula
Figure RE-GDA0003204350280000114
Determination of N0Representing the Gaussian white noise interference of said channel, wherein N represents the assigned channel set, function
Figure RE-GDA0003204350280000115
Representing a binary indicator function, when the variable X takes the value true,
Figure RE-GDA0003204350280000116
if not, then,
Figure RE-GDA0003204350280000117
collection
Figure RE-GDA0003204350280000118
Representing the set of terminals of the Internet of things distributed to the channel n ', k representing any one terminal of the Internet of things distributed to the channel n', hk，n′Representing the channel power gain, P, of the terminal k of the Internet of things in a channel nk，nRepresenting the transmitting power of the terminal k of the Internet of things in the channel n', and collecting
Figure RE-GDA0003204350280000119
Represents a set of terminals of the internet of things allocated to a channel n, and j represents any terminal m of the internet of things which is not equal to the terminal m of the internet of things and allocated to the channel nA terminal of Internet of things hj，nRepresenting the channel power gain, P, of the terminal j of the Internet of things in the channel nj，nRepresenting the transmitting power of the terminal j of the Internet of things in the channel n;
s303: the base station updates the initialized channel allocation result according to the data uploading delay to obtain an updated channel allocation result;
s304: and the base station distributes the selected channels in the channel set for each Internet of things terminal in the terminal set according to the updated channel distribution result.
Example 4
In this embodiment, on the basis of embodiments 1, 2 and 3, the step S303 is further expanded.
Further, as shown in fig. 5, the step S303 includes:
s401: the base station selects the Internet of things terminal m' with the largest data uploading delay in the terminal set;
s402: the base station searches a new channel n 'in the selected channel set for the terminal m' of the Internet of things;
s403: the base station judges whether the new channel n' simultaneously satisfies a condition a and a condition b:
condition a: the new channel n ' is a channel with the smallest data uploading delay of the internet of things terminal m ' in the selected channel set except for an initialized distribution channel n ', and the initialized distribution channel n ' is a channel distributed by the internet of things terminal m ' in the initialized channel distribution result;
condition b: the data uploading delay of the internet of things terminal m 'on the new channel n' is lower than that of the internet of things terminal m 'on the initialized distribution channel n';
if yes, executing step S404, otherwise executing step S405;
s404: updating the initialized distribution channel n 'to a new channel n';
s405: and taking the initialized channel allocation result as an updated channel allocation result.
It should be understood that the above-mentioned embodiments of the present invention are only examples for clearly illustrating the technical solutions of the present invention, and are not intended to limit the specific embodiments of the present invention. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention claims should be included in the protection scope of the present invention claims.

Claims (10)

1.一种网络边缘机器学习训练方法，其特征在于，包括以下步骤：1. a network edge machine learning training method, is characterized in that, comprises the following steps: 边缘服务器接收机器学习训练任务，读取所述机器学习训练任务的任务信息；The edge server receives the machine learning training task, and reads the task information of the machine learning training task; 所述边缘服务器根据所述任务信息确定数据源集合，所述数据源集合中一个数据源对应所述边缘服务器所在边缘网络内的一台物联网终端；The edge server determines a data source set according to the task information, and one data source in the data source set corresponds to an IoT terminal in the edge network where the edge server is located; 所述边缘服务器向所确定的数据源集合所对应的物联网终端发送数据采集请求；The edge server sends a data collection request to the IoT terminal corresponding to the determined data source set; 所述物联网终端接收所述数据采集请求，根据所述数据采集请求向所在的边缘网络的基站发送信道分配请求；The IoT terminal receives the data collection request, and sends a channel allocation request to the base station of the edge network where it is located according to the data collection request; 所述基站接收所述信道分配请求，根据所述信道分配请求为所述物联网终端分配信道；The base station receives the channel allocation request, and allocates a channel to the IoT terminal according to the channel allocation request; 所述物联网终端通过所分配的信道将本地数据集发送给所述边缘服务器；The IoT terminal sends the local data set to the edge server through the allocated channel; 所述边缘服务器接收所述本地数据集，根据所述本地数据集执行机器学习训练任务，返回训练结果。The edge server receives the local data set, performs a machine learning training task according to the local data set, and returns a training result. 2.根据权利要求1所述的一种网络边缘机器学习训练方法，其特征在于，所述任务信息包括任务所需数据类型；2. A kind of network edge machine learning training method according to claim 1, is characterized in that, described task information comprises task required data type; 所述边缘服务器根据所述任务信息确定数据源集合，包括：The edge server determines a data source set according to the task information, including: 所述边缘服务器统计所述边缘服务器所在边缘网络的所有物联网终端，以及各个物联网终端所采集数据类型；The edge server counts all IoT terminals in the edge network where the edge server is located, and the types of data collected by each IoT terminal; 所述边缘服务器选择所述采集数据类型与所述任务所需数据类型相同的物联网终端，根据所选择的所述物联网终端确定所述数据源集合。The edge server selects an IoT terminal whose type of the collected data is the same as the data type required by the task, and determines the data source set according to the selected IoT terminal. 3.根据权利要求1所述的一种网络边缘机器学习训练方法，其特征在于，所述边缘服务器向所确定的数据源集合所对应的物联网终端发送数据采集请求，包括：3. A network edge machine learning training method according to claim 1, wherein the edge server sends a data collection request to the Internet of Things terminal corresponding to the determined data source set, comprising: 所述边缘服务器根据所接收的所述机器学习训练任务，产生机器学习训练任务识别号；The edge server generates a machine learning training task identification number according to the received machine learning training task; 所述边缘服务器根据所述机器学习训练任务识别号，生成包括所述机器学习训练任务识别号的数据采集请求，向所确认的数据源集合所对应的物联网终端发送所述数据采集请求；The edge server generates a data collection request including the machine learning training task identification number according to the machine learning training task identification number, and sends the data collection request to the IoT terminal corresponding to the confirmed data source set; 所述物联网终端接收所述数据采集请求，根据所述数据采集请求向所在的边缘网络的基站发送信道分配请求，包括：The IoT terminal receives the data collection request, and sends a channel allocation request to the base station of the edge network where it is located according to the data collection request, including: 所述物联网终端接收所述数据采集请求，根据所述数据采集请求生成包括所述机器学习训练任务识别号的信道分配请求，向所在的边缘网络的基站发送所述信道分配请求；The IoT terminal receives the data collection request, generates a channel allocation request including the machine learning training task identification number according to the data collection request, and sends the channel allocation request to the base station of the edge network where it is located; 所述基站接收所述信道分配请求，根据所述信道分配请求为所述物联网终端分配信道，包括：The base station receives the channel allocation request, and allocates a channel to the IoT terminal according to the channel allocation request, including: 所述基站接收所述信道分配请求，将所接收的所述信道分配请求中所述机器学习训练任务识别号相同的所述物联网终端形成终端集合；The base station receives the channel allocation request, and forms a terminal set from the IoT terminals with the same machine learning training task identification number in the received channel allocation request; 所述基站根据所述信道分配请求中的所述机器学习训练任务识别号选取对应的信道集合，不同的所述机器学习训练任务识别号对应不同的所述信道集合；The base station selects a corresponding channel set according to the machine learning training task identification number in the channel allocation request, and different machine learning training task identification numbers correspond to different channel sets; 所述基站为所述终端集合中的各个物联网终端，分配所选取的所述信道集合中的信道。The base station allocates the selected channel in the channel set to each IoT terminal in the terminal set. 4.根据权利要求3所述的一种网络边缘机器学习训练方法，其特征在于，4. a kind of network edge machine learning training method according to claim 3, is characterized in that, 所述基站为所述终端集合中的各个物联网终端，分配所选取的所述信道集合中的信道，包括：The base station allocates the selected channels in the channel set for each IoT terminal in the terminal set, including: 所述基站根据所选取的所述信道集合中的信道，对所述终端集合中各个物联网终端进行初始化信道分配，得到初始化信道分配结果；The base station performs initial channel allocation for each IoT terminal in the terminal set according to the selected channel in the channel set, and obtains an initial channel allocation result; 所述基站根据所述初始化信道分配结果，计算所述终端集合中各个物联网终端在不同分配信道上的数据上传延迟，各个物联网终端在不同分配信道上的数据上传延迟为各个所述物联网终端通过不同分配信道将本地数据集发送给所述边缘服务器的延迟；The base station calculates the data upload delay of each IoT terminal on different allocation channels in the terminal set according to the initialized channel allocation result, and the data upload delay of each IoT terminal on different allocation channels is the same as that of each IoT terminal. The delay of the terminal sending the local data set to the edge server through different distribution channels; 所述基站根据所述数据上传延迟对所述初始化信道分配结果进行更新，得到更新后信道分配结果；The base station updates the initialized channel allocation result according to the data upload delay to obtain the updated channel allocation result; 所述基站根据所述更新后信道分配结果为所述终端集合中的各个物联网终端，分配所选取的所述信道集合中的信道。The base station allocates the selected channel in the channel set to each IoT terminal in the terminal set according to the updated channel allocation result. 5.根据权利要求4所述的一种网络边缘机器学习训练方法，其特征在于，所述基站根据所选取的所述信道集合中的信道，对所述终端集合中各个物联网终端进行初始化信道分配，得到初始化信道分配结果，包括：5 . The network edge machine learning training method according to claim 4 , wherein the base station initializes a channel for each IoT terminal in the terminal set according to the selected channel in the channel set. 6 . Assign, get the initial channel assignment results, including: 所述基站计算所述终端集合中各个物联网终端对应不同信道的信道功率增益，对应不同信道的所述信道功率增益为所述物联网终端通过所述信道集合中的不同信道，将本地数据集发送给所述边缘服务器的功率增益；The base station calculates the channel power gains corresponding to different channels of each IoT terminal in the terminal set, and the channel power gains corresponding to the different channels are obtained by the IoT terminal using the different channels in the channel set to convert the local data set. a power gain sent to the edge server; 所述基站将各个物联网终端信道功率增益最大时所对应的信道分配给所述物联网终端，得到初始化信道分配结果。The base station allocates the channel corresponding to the maximum channel power gain of each IoT terminal to the IoT terminal, and obtains an initialized channel allocation result. 6.根据权利要求4所述的一种网络边缘机器学习训练方法，其特征在于，所述基站根据所述初始化信道分配结果，计算所述终端集合中各个物联网终端在不同分配信道上的数据上传延迟，包括：6 . The network edge machine learning training method according to claim 4 , wherein the base station calculates the data of each IoT terminal in the terminal set on different allocation channels according to the initialized channel allocation result. 7 . Upload delays, including: 所述基站采用公式Dm＝Lm/Rm，n计算所述终端集合中各个物联网终端的数据上传延迟，其中，Lm表示所述信道分配物联网终端集合中任意一台物联网终端m的本地数据集的大小，Rm，n表示所述物联网终端m在所分配信道n中的数据上传速率。The base station uses the formula D m =L m /R m,n to calculate the data upload delay of each IoT terminal in the terminal set, where L m represents any IoT terminal in the channel allocation IoT terminal set The size of the local data set of m, R m, n represents the data upload rate of the IoT terminal m in the allocated channel n. 7.根据权利要求4所述的一种网络边缘机器学习训练方法，其特征在于，所述基站根据所述数据上传延迟对所述初始化信道分配结果进行更新，得到更新后信道分配结果，包括：7. The network edge machine learning training method according to claim 4, wherein the base station updates the initialized channel allocation result according to the data upload delay, and obtains the updated channel allocation result, comprising: 所述基站选择所述终端集合中数据上传延迟最大的物联网终端m′；The base station selects the IoT terminal m' with the largest data upload delay in the terminal set; 所述基站为所述物联网终端m′在所选取的所述信道集合中寻找新信道n″；The base station searches for a new channel n" in the selected channel set for the IoT terminal m'; 所述基站判断所述新信道n″是否同时满足条件a和条件b：The base station determines whether the new channel n" satisfies both condition a and condition b: 条件a：所述新信道n″为所选取的所述信道集合中的，除了初始化分配信道n′以外的，所述物联网终端m′的数据上传延迟最小的信道，所述初始化分配信道n′为所述初始化信道分配结果中所述物联网终端m′所分配的信道；Condition a: The new channel n" is the channel in the selected channel set, except for the initialized allocation channel n', the channel with the smallest data upload delay of the IoT terminal m', the initialized allocation channel n ' is the channel allocated by the IoT terminal m' in the initialized channel allocation result; 条件b：所述物联网终端m′在所述新信道n″上的数据上传延迟低于所述物联网终端m′在所述初始化分配信道n′上的数据上传延迟；Condition b: the data upload delay of the IoT terminal m' on the new channel n" is lower than the data upload delay of the IoT terminal m' on the initial allocation channel n'; 若是，则将初始化分配信道n′更新为新信道n″，继续执行计算所述终端集合中各个物联网终端在不同分配信道上的数据上传延迟的步骤，若否，则将所述初始化信道分配结果作为更新后的信道分配结果。If so, update the initialized allocation channel n' to a new channel n", and continue to perform the step of calculating the data upload delay of each IoT terminal in the terminal set on different allocation channels; if not, allocate the initialized channel The result is used as the updated channel assignment result. 8.根据权利要求2所述的一种网络边缘机器学习训练方法，其特征在于，所述任务信息还包括任务类型、任务所采用的机器学习算法以及任务精度要求；8. A network edge machine learning training method according to claim 2, wherein the task information further comprises a task type, a machine learning algorithm adopted by the task and a task precision requirement; 所述边缘服务器接收所述本地数据集，根据所述本地数据集执行机器学习训练任务，返回训练结果，包括：The edge server receives the local data set, performs a machine learning training task according to the local data set, and returns a training result, including: 所述边缘服务器接收所述本地数据集；the edge server receives the local dataset; 所述边缘服务器将所述本地数据集进行汇聚；the edge server aggregates the local data set; 所述边缘服务器以所汇聚的所述本地数据集为数据源，根据所述任务类型、任务所采用的机器学习算法执行机器学习训练；The edge server uses the aggregated local data set as a data source, and performs machine learning training according to the task type and the machine learning algorithm adopted by the task; 所述边缘服务器判断所述机器学习训练的精度是否不低于所述任务精度要求，若是，保存训练结果，结束训练，若否，继续执行以所汇聚的所述本地数据集为数据源，根据所述任务类型、任务所采用的机器学习算法执行机器学习训练步骤。The edge server judges whether the accuracy of the machine learning training is not lower than the task accuracy requirement, and if so, saves the training result, and ends the training, if not, continues to execute the collected local data set as the data source, according to The task type and the machine learning algorithm adopted by the task perform the machine learning training step. 9.根据权利要求8所述的一种网络边缘机器学习训练方法，其特征在于，所述任务信息还包括任务可容忍的最大延迟；9. A network edge machine learning training method according to claim 8, wherein the task information further comprises the maximum delay that the task can tolerate; 在边缘服务器接收机器学习训练任务，读取所述机器学习训练任务的任务信息之后，还包括：所述边缘服务器根据所述任务可容忍的最大延迟启动机器学习训练任务计时器，所述机器学习训练任务计时器以所述任务可容忍的最大延迟为阈值；After the edge server receives the machine learning training task and reads the task information of the machine learning training task, the method further includes: the edge server starts the machine learning training task timer according to the maximum delay that can be tolerated by the task, and the machine learning The training task timer is thresholded by the maximum delay that the task can tolerate; 在所述边缘服务器以所汇聚的所述本地数据集为数据源，根据所述任务类型、任务所采用的机器学习算法执行机器学习训练之后，以及判断模型训练的精度是否不低于所述模型训练的精度要求之前，还包括：After the edge server uses the aggregated local data set as a data source, performs machine learning training according to the task type and the machine learning algorithm used in the task, and determines whether the accuracy of model training is not lower than that of the model Before training accuracy requirements, also include: 判断所述机器学习训练任务计时器是否超时，若是，保存训练结果，结束训练，若否，继续执行判断模型训练的精度是否不低于所述模型训练的精度要求。Determine whether the machine learning training task timer has timed out, if so, save the training result and end the training, if not, continue to judge whether the accuracy of the model training is not lower than the accuracy requirement of the model training. 10.根据权利要求9所述的一种网络边缘机器学习训练方法，其特征在于，在所述边缘服务器向所确认的数据源集合所对应的物联网终端发送数据采集请求之后，还包括：所述边缘服务器启动数据采集计时器，所述数据采集计时器的阈值小于所述机器学习训练任务计时器的阈值；10. A network edge machine learning training method according to claim 9, wherein after the edge server sends a data collection request to the IoT terminal corresponding to the confirmed data source set, the method further comprises: The edge server starts a data collection timer, and the threshold of the data collection timer is smaller than the threshold of the machine learning training task timer; 在所述边缘服务器接收所述本地数据集之后，以及所述边缘服务器将所述本地数据集进行汇聚之前，包括：After the edge server receives the local data set and before the edge server aggregates the local data set, it includes: 所述边缘服务器判断是否已接收到数据源集合中对应的所有物联网终端的本地数据集，若是，则继续执行所述边缘服务器将所述本地数据集进行汇聚步骤，若否，继续判断所述数据采集计时器是否超时，若是，则继续执行所述边缘服务器将所述本地数据集进行汇聚步骤，若否，则返回执行所述边缘服务器接收所述本地数据集。The edge server judges whether the local data sets of all IoT terminals corresponding to the data source set have been received, and if so, continues to perform the step of aggregating the local data sets by the edge server. Whether the data collection timer times out, if yes, continue to perform the step of aggregating the local data set by the edge server, if not, return to the edge server to receive the local data set.
CN202110897051.9A 2021-08-05 2021-08-05 Network edge machine learning training method Pending CN113692052A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110897051.9A CN113692052A (en) 2021-08-05 2021-08-05 Network edge machine learning training method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110897051.9A CN113692052A (en) 2021-08-05 2021-08-05 Network edge machine learning training method

Publications (1)

Publication Number Publication Date
CN113692052A true CN113692052A (en) 2021-11-23

Family

ID=78578925

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110897051.9A Pending CN113692052A (en) 2021-08-05 2021-08-05 Network edge machine learning training method

Country Status (1)

Country Link
CN (1) CN113692052A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114401519A (en) * 2022-02-18 2022-04-26 广东技术师范大学 Automatic construction method of underwater three-dimensional wireless sensor network
CN116644802A (en) * 2023-07-19 2023-08-25 支付宝(杭州)信息技术有限公司 Model training method and device

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107734558A (en) * 2017-10-26 2018-02-23 北京邮电大学 A kind of control of mobile edge calculations and resource regulating method based on multiserver
CN111212108A (en) * 2019-12-12 2020-05-29 西安电子科技大学 Multi-user parallel migration method based on non-orthogonal multiple access and mobile edge computing
CN113098806A (en) * 2021-04-16 2021-07-09 华南理工大学 Method for compressing cooperative channel adaptability gradient of lower end in federated learning

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107734558A (en) * 2017-10-26 2018-02-23 北京邮电大学 A kind of control of mobile edge calculations and resource regulating method based on multiserver
CN111212108A (en) * 2019-12-12 2020-05-29 西安电子科技大学 Multi-user parallel migration method based on non-orthogonal multiple access and mobile edge computing
CN113098806A (en) * 2021-04-16 2021-07-09 华南理工大学 Method for compressing cooperative channel adaptability gradient of lower end in federated learning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
MIAN GUO;CHUN SHAN;MITHUN MUKHERJEE;JAIME LLORET;QUANSHENG GUAN: "《Collaborative Edge Learning in MIMO-NOMA Uplink Transmission Environment》", 《2021 IEEE/CIC INTERNATIONAL CONFERENCE ON COMMUNICATIONS IN CHINA (ICCC)》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114401519A (en) * 2022-02-18 2022-04-26 广东技术师范大学 Automatic construction method of underwater three-dimensional wireless sensor network
CN114401519B (en) * 2022-02-18 2023-06-09 广东技术师范大学 A method for automatic construction of underwater three-dimensional wireless sensor network
CN116644802A (en) * 2023-07-19 2023-08-25 支付宝(杭州)信息技术有限公司 Model training method and device

Similar Documents

Publication Publication Date Title
Prathiba et al. Federated learning empowered computation offloading and resource management in 6G-V2X
You et al. Asynchronous mobile-edge computation offloading: Energy-efficient resource management
CN109814951B (en) Joint optimization method for task unloading and resource allocation in mobile edge computing network
WO2021233053A1 (en) Computing offloading method and communication apparatus
WO2019200716A1 (en) Fog computing-oriented node computing task scheduling method and device thereof
CN110087318A (en) Task unloading and resource allocation joint optimization method based on the mobile edge calculations of 5G
CN112105062B (en) Mobile edge computing network energy consumption minimization strategy method under time-sensitive condition
CN113419857A (en) Federal learning method and system based on edge digital twin association
CN113645273B (en) Internet of vehicles task unloading method based on service priority
CN111511028B (en) Multi-user resource allocation method, device, system and storage medium
CN110753319B (en) Heterogeneous service-oriented distributed resource allocation method and system in heterogeneous Internet of vehicles
CN109756912B (en) Multi-user multi-base station joint task unloading and resource allocation method
Kopras et al. Task allocation for energy optimization in fog computing networks with latency constraints
CN111246586A (en) A method and system for allocating smart grid resources based on genetic algorithm
Younis et al. Energy-latency-aware task offloading and approximate computing at the mobile edge
CN114173421B (en) LoRa logical channel and power allocation method based on deep reinforcement learning
WO2024113974A1 (en) Computing power network routing allocation method and apparatus, electronic device, and storage medium
CN114827191B (en) Dynamic task unloading method for fusing NOMA in vehicle-road cooperative system
EP4383075A1 (en) Data processing method and apparatus
CN113692052A (en) Network edge machine learning training method
KR102298698B1 (en) Method and apparatus for service caching in edge computing network
Paymard et al. Task scheduling based on priority and resource allocation in multi-user multi-task mobile edge computing system
CN117851356A (en) A UAV-assisted caching strategy method and system based on deep Q network
WO2024070057A1 (en) Admission control method and admission request method in communication network system
WO2022082742A1 (en) Model training method and device, server, terminal, and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20211123

WD01 Invention patent application deemed withdrawn after publication