CN113315978B - Collaborative online video edge caching method based on federal learning - Google Patents
Collaborative online video edge caching method based on federal learning Download PDFInfo
- Publication number
- CN113315978B CN113315978B CN202110521197.3A CN202110521197A CN113315978B CN 113315978 B CN113315978 B CN 113315978B CN 202110521197 A CN202110521197 A CN 202110521197A CN 113315978 B CN113315978 B CN 113315978B
- Authority
- CN
- China
- Prior art keywords
- user
- users
- online video
- edge
- prediction model
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 89
- 238000012549 training Methods 0.000 claims abstract description 62
- 230000004044 response Effects 0.000 claims description 22
- 238000012545 processing Methods 0.000 claims description 9
- 230000001419 dependent effect Effects 0.000 claims description 4
- 230000002238 attenuated effect Effects 0.000 claims description 3
- 230000001413 cellular effect Effects 0.000 claims description 3
- 230000008569 process Effects 0.000 description 7
- 238000004891 communication Methods 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 4
- 230000007423 decrease Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 230000001537 neural effect Effects 0.000 description 3
- 230000002776 aggregation Effects 0.000 description 2
- 238000004220 aggregation Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000012952 Resampling Methods 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
- H04N21/21—Server components or server architectures
- H04N21/218—Source of audio or video content, e.g. local disk arrays
- H04N21/2181—Source of audio or video content, e.g. local disk arrays comprising remotely distributed storage units, e.g. when movies are replicated over a plurality of video servers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
- G06F16/73—Querying
- G06F16/735—Filtering based on additional data, e.g. user or group profiles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/906—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/957—Browsing optimisation, e.g. caching or content distillation
- G06F16/9574—Browsing optimisation, e.g. caching or content distillation of access to content, e.g. by caching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
- H04N21/23—Processing of content or additional data; Elementary server operations; Server middleware
- H04N21/231—Content storage operation, e.g. caching movies for short term storage, replicating data over plural servers, prioritizing data for deletion
- H04N21/23106—Content storage operation, e.g. caching movies for short term storage, replicating data over plural servers, prioritizing data for deletion involving caching operations
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
- H04N21/23—Processing of content or additional data; Elementary server operations; Server middleware
- H04N21/234—Processing of video elementary streams, e.g. splicing of video streams or manipulating encoded video stream scene graphs
- H04N21/23406—Processing of video elementary streams, e.g. splicing of video streams or manipulating encoded video stream scene graphs involving management of server-side video buffer
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Signal Processing (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)
- Information Transfer Between Computers (AREA)
Abstract
The invention discloses a collaborative online video edge caching method based on federal learning, wherein a participating object comprises a plurality of users, a plurality of edge nodes and a central server, the plurality of edge nodes provide services for the plurality of movable users in the coverage area of the edge nodes and are connected to the central server, and each edge node is provided with an edge server, the method comprises the following steps: step one, establishing a network model according to a plurality of users, a plurality of edge nodes and a central server of participating objects; step two, establishing a prediction model by adopting an improved federal learning method and training the prediction model to obtain a user online video request prediction model; and step three, obtaining a user online video request prediction list according to the user online video request prediction model, analyzing the user request prediction list in the area coverage range by a plurality of edge nodes, and performing edge caching by adopting a collaborative caching decision. According to the method, the prediction model after training can be automatically updated.
Description
Technical Field
The invention relates to a collaborative video edge caching method based on federal learning, and belongs to the technical field of edge caching of cloud edge collaboration.
Background
With the development of mobile internet social platforms and the popularization of intelligent terminal devices, people's demand for high-quality real-time data rises sharply, especially for video services such as short videos and live broadcasts. The common mode of acquiring user requirements is used for training a prediction model of a user online video request in a cloud computing center, the traditional cloud computing mode requires the cloud center to have strong computing and storage capacities, a large amount of user request information needs to be transmitted to the cloud center in the training process, the transmission process not only occupies network flow, but also can cause privacy disclosure of users. If the use period of the prediction model is selected to be prolonged in order to reduce communication with the cloud center, the prediction model may be outdated, and the prediction result is inaccurate. Further, with the arrival of the 5G era, the traditional cloud computing mode cannot meet the requirement of low delay of users, and the 5G network also increases the load of the backhaul link.
In order to make up for the defects of the cloud computing mode, researchers put forward the concept of edge computing and cloud edge coordination, and consider caching the video by using the mode of edge caching, so that the requirement of a user on low time delay is met, and the backhaul flow is reduced. The core of the edge caching technology is how to make full use of the storage resources of each edge node and place contents with a certain degree of demand on the edge nodes close to users. Because the edge node usually has a limited storage capacity, it is difficult to select and cache contents with higher user requirements from a large amount of contents in a limited storage space.
In the prior art, a prediction model is mostly established through various different-depth learning methods, online video requests of users are predicted and sequenced, and then cached contents are sequentially selected. However, these methods using deep learning also require the user's historical request data to be sent to the cloud center before the predictive model can be trained. In the process of sending the data to the cloud center, the risk of revealing user privacy data exists, and a large amount of broadband waste resources are occupied; moreover, because the user's preference is time-sensitive, the trained model may be too old, and the user request data needs to be collected continuously to retrain the failed prediction model.
Saputra et al propose to predict content popularity based on a distributed deep learning framework, distribute computing tasks generated by users to multiple servers for computing processing, and the specific way of performing deep learning depends on the proposed deep learning model, such as convolution. The distributed learning model requires that data partitioning on different computing nodes is independent and distributed, but in order to protect privacy, users generally do not want to leave the local area for requesting data, so that a data set generated by each user application is more personalized and is not independent and distributed. Therefore, the ideal data requirement based on the distributed deep learning model cannot meet the scene of practical application.
Disclosure of Invention
For the edge cache of the online video, in order to establish a more accurate user online video request prediction model and select the content with higher user demand degree for caching, thereby improving the hit rate of the online video cache, reducing the user waiting time delay, reducing the return flow and simultaneously reducing the communication cost and the caching cost, the invention provides a collaborative online video edge cache method based on federal learning by combining the characteristic that an edge cache scene deploys a plurality of edge nodes. Meanwhile, the invention also provides a scheme of cache decision and service response with cooperative consciousness among a plurality of edge nodes, reduces the storage of a large amount of repeated contents in adjacent areas, saves the cache cost and further reduces the time delay and the return flow.
The invention discloses a collaborative video edge caching method based on federal learning, which has the following specific technical scheme:
a collaborative online video edge caching method based on federal learning, wherein a participating object comprises a plurality of users, a plurality of edge nodes and a central server, the plurality of edge nodes provide services for the plurality of users movable in different geographical positions in the coverage area of the edge nodes through a wireless cellular network and are connected to the central server through a backhaul link, and each edge node is provided with an edge server with computing capability and storage capability, the method comprises the following steps:
step one, establishing a network model according to a plurality of users, a plurality of edge nodes and a central server of participating objects;
step two, establishing a prediction model by adopting an improved federal learning method, training the prediction model to obtain a user online video request prediction model, and predicting the user online video request;
and step three, obtaining a user online video request prediction list according to the user online video request prediction model, analyzing the user request prediction list in the area coverage range by a plurality of edge nodes, and performing edge caching by adopting a collaborative caching decision.
The collaborative online video edge caching method based on the federal learning further comprises the step of automatically updating the prediction model after the training in the step two.
According to the collaborative online video edge caching method based on federal learning, optionally, the method for automatically updating the prediction model comprises the following steps: and the edge node records the service response condition of the online video request of the user, calculates the hit rate of the edge cache and the hit rate of the cooperative cache, compares the calculated result with a set threshold, and trains the model again to realize automatic updating if the calculated result is lower than the threshold.
In one embodiment, for ease of description,
for received online video requests, xfE {0,1} represents the cache state, x f1 means there is a buffer, otherwise there is no buffer. The edge node records the service response condition of the user request and calculates the cache hit rate
And cooperative cache hit rate
And comparing the calculated index with a set threshold, if the calculated index is lower than the threshold, the model is out of date and needs to be trained again.
According to the collaborative online video edge caching method based on the federal learning, optionally, the central server can also be in direct communication with a plurality of users, and the specific steps of establishing a prediction model for the online video request of the user by adopting the improved federal learning method in the step two are as follows:
s1, a central server distributes a prediction model training instruction to a user, wherein the prediction model training instruction comprises a method for processing data of request data of the user and selection of a prediction model;
s2, each user carries out embedding processing on user information and video information in local request data of the user and sends the embedded user information to a central server;
s3, clustering the embedded user information by the central server to form a p-type user set; extracting m users from the p types of user sets to participate in model training, ensuring that the number of the users randomly extracted from each type of user set is equal, and instructing the m users to train a prediction model;
s4, initializing global prediction model parameters by the central server and distributing the global prediction model parameters to the extracted m users, locally training the prediction model by each user by using local request data, and sending the trained local model parameters to the central server;
s5, the central server aggregates local model parameters of m users after local training to obtain updated global model parameters, and sends the updated global model parameters to all the users to update the parameters in a new round;
and S6, repeating the steps of S2-S5 until the model converges to obtain a prediction model of the user online video request, and performing request prediction to obtain a prediction list.
According to the collaborative online video edge caching method based on the federal learning, optionally, when the prediction model is subjected to parameter updating, an improved federal learning parameter updating method is adopted to perform parameter updating on the user prediction model.
According to the collaborative online video edge caching method based on the federal learning, optionally, in an improved federal learning parameter updating method, clustered similar users are collected, users which are extracted to participate in model training are used as representative users, and other users which are not extracted are used as subordinate users; and updating the parameters of the prediction model of the subordinate user when a next round of model training is needed after the representative user completes the local training and the central server completes the global parameter updating of the prediction model by utilizing the similarity of the similar users after clustering.
According to the collaborative online video edge caching method based on the federal learning, parameters needing to be updated in a user prediction model are divided into user information embedding parameters, video information embedding parameters and non-embedding parameters, and the parameters are updated by adopting the following rules:
aiming at the user information embedding parameters, in the similar users, the representative user adopts the embedding parameter value updated after local training, and the subordinate user adopts the embedding parameter value of the subordinate user in the previous round and the average value of the variation of the user information embedding parameters of all representative users in the similar user set attenuated along with the increase of the round;
for the video information embedding parameter, the embedding parameter of the video information directly related to the representative user is updated to be the influence factor of the representative user on the video, multiplied by the video information embedding parameter of the corresponding representative user and divided by the sum of the influence factor of each representative user on the video;
aiming at non-embedded parameters, a traditional parameter updating method of federal learning is adopted.
In one embodiment, the improved federal learning parameter update method employs the following parameter update rules:
nka number of training data representing user k; c represents a set of user groups; ckA user set representing a user group in which a user k is located; s represents a user set which is extracted to participate in the current round of training; w represents a parameter matrix, w [ J ]]Representing a partial parameter matrix corresponding to the index J; u represents the index of the user embedding parameter, and I represents the index of the online video embedding parameter; u shapekAn index representing user k embedding parameters; i iskAn index of embedded parameters representing the online video in which user k is associated; n denotes the index of the non-embedded parameter.
In the same-class user set, the users extracted to participate in the training of the current round are taken as representative users, and the other users not extracted are taken as subordinate users, so that the updating mode of the model parameters is as follows:
for the user information embedding parameter representing the user, enabling the representing user to belong to the S, and adopting the embedding parameter value updated after local training by the embedding parameter updating rule, namely:
wt+1[Uk]＝wt[Uk]
for the user information embedding parameters of the subordinate users, because the users in the same user group have certain similarity after clustering operation, the embedding parameters of other subordinate users are related to the updating of the embedding parameters of the representative user extracted to participate in training in the user group, so that the representative user k belongs to S, and the subordinate user C belongs to CkS, the sum of the variance of the user embedding parameters of all the representative users in the user group is delta Uc]＝Σk∈S(wt[Uk]-wt-1[Uk]) The total number of representative users of the user group is | CkAnd n S, γ is the discount rate, and decreases with the increase of the training round, and the dependent user embedding parameter updating rule adopts the original embedding parameter plus the mean value of the variation of the user information embedding parameters of all the representative users of the user group, which attenuates with the increase of the training round, that is:
for the embedded parameters of the video information, the video request is related to the user, the embedded parameters of the video information directly related to the representative user can be correspondingly updated, and the influence factors of the representative user Kepsilon S and the user k on all the related videos are enabled to be
Determined by the sum of the variations before and after updating of its embedding parameters, i epsilon U per videokεSIkThen, the video information embedding parameter updating rule adopts the influence factor of the representative user k on the video i multiplied by the video information embedding parameter of the corresponding representative user and divided by the sum of the influence factors of each representative user on the video, that is:
for non-embedded parameters, adopting an update rule in the traditional federal learning to enable a representative user k to be belonged to S and nσ＝∑k∈SnkThe update rule of the non-embedded parameters is the weighted average of the whole of all the representative users, i.e. the total number of data trained for all the representative users
According to the federally learned collaborative online video edge caching method, preferably, the method for performing collaborative caching decision by a plurality of edge nodes is as follows:
a1 obtaining a prediction list of a plurality of user online video requests in the coverage area range of each edge node;
a2, calculating the caching value of the online video according to the occurrence frequency of different online videos in the prediction lists of a plurality of users and the participation of the users;
a3 according to the online video caching value, the edge servers of each edge node cooperate to cache until there is no redundant storage space.
According to the collaborative online video edge caching method based on the federal learning, optionally, a plurality of edge nodes carry out collaborative service response on online video requests of users.
According to the collaborative online video edge caching method based on the federal learning, optionally, the collaborative service response includes: when each edge node receives a user online video request in the coverage area of the edge node, the edge node judges whether the online video exists in a cache list of the current node or not: if the local service exists, the edge node carries out local service response, otherwise, the edge node caches the neighbor nodes; and judging whether the online video exists in a cache list of the neighbor node: if the neighbor node exists, the neighbor node performs service response, otherwise, the central server performs service response.
The invention has the beneficial effects that:
according to the collaborative online video edge caching method based on the federal learning, an improved federal learning method is adopted in the process of establishing a prediction model of a user online video request, users are clustered firstly, and then user request data are sampled, so that the training speed and the model convergence speed of the prediction model can be greatly improved, and the communication cost is reduced.
Because each user carries out a certain round of local training respectively in the traditional federal learning training process, then the respective parameter updating results are transmitted to the center for aggregation, the center aggregates the parameters after overall updating and distributes the parameters to each user for parameter updating, the traditional federal learning parameter updating method only updates the extracted user parameters each time, which leads the user clustering to tend to be unchanged after few rounds, but the improved federal learning method adopted by the invention increases user clustering resampling, if the method (weighted average) of the traditional federal learning center aggregation parameters is adopted, namely, the operation of user clustering is ineffective afterwards, therefore, the invention also provides a pertinently improved federal learning parameter updating method according to the similarity of the similar users after clustering, and updates the subordinate users which are not extracted in the similar user set according to the extracted parameters of the representative users, the effectiveness of the clustering operation is maintained.
Furthermore, due to the change of time, the demand of the online video request of the user changes along with the change of time, so that the trained prediction model related to the user request is outdated, the prediction accuracy is reduced, the high hit rate of the online video of the edge cache cannot be ensured, the waiting time delay of the user is increased, and the storage resources are wasted.
According to the collaborative online video edge caching method based on the federal learning, the prediction model of the online video request of the user is obtained through the improved federal learning method, the effectiveness of the prediction model is automatically detected according to the feedback, and the prediction model is automatically updated, so that the privacy of the user can be protected, and the communication cost for transmitting mass user request information in the process of training the prediction model is reduced on the premise of ensuring the prediction accuracy. Furthermore, by adopting a cooperative cache decision and service response scheme, the invention can obviously improve the cache hit rate, reduce the user waiting time delay and the return flow and reduce the cache cost.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a diagram of a network model architecture of an edge caching method according to the present invention;
FIG. 2 is a flow chart of an improved federated learning parameter update method in an edge caching method according to the present invention;
FIG. 3 is a flow chart of an edge caching method according to the present invention;
FIG. 4 is a schematic diagram of predictive model training in the edge caching method according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
The first embodiment is as follows:
the embodiment provides a collaborative video edge caching method based on federal learning, and the collaborative video edge caching method based on federal learning according to the embodiment has the following specific technical scheme:
a collaborative online video edge caching method based on federal learning, wherein a participating object comprises a plurality of users, a plurality of edge nodes and a central server, the plurality of edge nodes provide services for the plurality of users movable in different geographical positions in the coverage area of the edge nodes through a wireless cellular network and are connected to the central server through a backhaul link, and each edge node is provided with an edge server with computing capability and storage capability, the method comprises the following steps:
step one, establishing a network model according to a plurality of users, a plurality of edge nodes and a central server of participating objects, and referring to fig. 1;
step two, establishing a prediction model by adopting an improved federal learning method, training the prediction model to obtain a user online video request prediction model, and predicting the user online video request;
and step three, obtaining a user online video request prediction list according to the user online video request prediction model, analyzing the user request prediction list in the area coverage range by a plurality of edge nodes, and performing edge caching by adopting a collaborative caching decision.
The collaborative online video edge caching method based on federated learning according to this embodiment also includes that the prediction model trained in step two is automatically updated.
According to the collaborative online video edge caching method based on federal learning in the embodiment, the method for automatically updating the prediction model comprises the following steps: and the edge node records the service response condition of the online video request of the user, calculates the hit rate of the edge cache and the hit rate of the cooperative cache, compares the calculated result with a set threshold, and trains the model again to realize automatic updating if the calculated result is lower than the threshold.
In the present embodiment, for convenience of description,
on-line video requests, x, received for edge nodesfE {0,1} represents the cache state of the edge node, xf1 means there is a buffer, otherwise there is noThere is a cache. The edge node records the service response condition of the user request and calculates the cache hit rate
And cooperative cache hit rate
And comparing the calculated index with a set threshold, if the calculated index is lower than the threshold, the model is out of date and needs to be trained again.
According to the collaborative online video edge caching method based on federal learning in the embodiment, in the network model, the central server can also directly communicate with a plurality of users, and the specific steps of establishing a prediction model for the online video request of the user by adopting an improved federal learning method in the step two are as follows:
s1, a central server distributes a prediction model training instruction to a user, wherein the prediction model training instruction comprises a method for processing data of request data of the user and selection of a prediction model;
s2, each user carries out embedding processing on user information and video information in local request data of the user and sends the embedded user information to a central server;
s3, clustering the embedded user information by the central server to form a p-type user set; extracting m users from the p types of user sets to participate in model training, ensuring that the number of the users randomly extracted from each type of user set is equal, and instructing the m users to train a prediction model;
s4, initializing global prediction model parameters by the central server and distributing the global prediction model parameters to the extracted m users, locally training the prediction model by each user by using local request data, and sending the trained local model parameters to the central server;
s5, the central server aggregates local model parameters of m users after local training to obtain updated global model parameters, and sends the updated global model parameters to all the users to update the parameters in a new round;
and S6, repeating the steps of S2-S5 until the model converges to obtain a prediction model of the user online video request, and performing request prediction to obtain a prediction list.
Referring to fig. 4, for the sake of detailed description, in the present embodiment, the neural collaborative filtering is selected, and a well-known recommendation algorithm is explained as a network model that needs to be trained, but the method and process of the present invention are not limited to the selection of the network model, and any selection and modification on the network model can be made by those skilled in the art after understanding the spirit of the present invention.
According to the selected algorithm, the training process of the prediction model of the embodiment is as follows:
s1, a central server distributes a prediction model training instruction to a user, wherein the prediction model training instruction comprises a method for processing data of request data of the user and selection of a prediction model;
s2, each user carries out embedding processing on user information and video information in local request data of the user, and sends embedded user information embedding parameters to a central server;
s3, clustering by the central server through the user information embedding parameters to form a p-type user set; extracting m users from the p types of user sets to participate in model training, ensuring that the number of the users randomly extracted from each type of user set is equal, and instructing the m users to train a prediction model;
s4, the central server initializes global prediction model parameters and distributes the global prediction model parameters to the extracted m users, the extracted m users carry out local training by utilizing own data of the users and send the local parameters after the training to the central server, wherein taking a neural filter collaborative network model as an example, a first layer of neural collaborative filter layer trained by the users is defined as phi1(pu,qi)＝pu⊙ qiBy, indicates a vector inner product, and the output layer indicates
aoutIs output layer activation letterThe number h is the weight of an output layer, and the parameters are updated through the calculation of a loss function;
and S5, integrating the local model parameters after the m users are trained by the central server to obtain updated global model parameters, sending the updated global model parameters to all the users, and updating the parameters in a new round, wherein as shown in FIG. 2, when the prediction model is updated, an improved federal learning parameter updating method is adopted to update the user request data.
In the improved federal learning parameter updating method, clustered similar users are concentrated, users which are extracted to participate in model training are used as representative users, and other users which are not extracted are used as subordinate users; and updating the parameters of the prediction model of the dependent user when a representative user needs to perform next round of model training after completing each round of local training and completing model global parameter updating by utilizing the similarity of the similar users after clustering.
According to the collaborative online video edge caching method based on federal learning in the embodiment, parameters needing to be updated in a user prediction model are divided into user information embedding parameters, video information embedding parameters and non-embedding parameters, and the parameters are updated by adopting the following rules:
aiming at the user information embedding parameters, in the same-class user set, representing users adopt the embedding parameter values updated after local training, and subordinate users adopt the embedding parameter values of the subordinate users in the previous round and the variation mean values of the user information embedding parameters of all the representing users in the same-class user set attenuated along with the increase of the round;
for the video information embedding parameter, the embedding parameter of the video information directly related to the representative user is updated to be the influence factor of the representative user on the video, multiplied by the video information embedding parameter of the corresponding representative user and divided by the sum of the influence factor of each representative user on the video;
aiming at non-embedded parameters, a traditional parameter updating method of federal learning is adopted.
In this embodiment, the improved federal learning parameter update method employs the following parameter update rule:
nka number of training data representing user k; c represents a set of user groups; ckA user set representing a user group in which a user k is located; s represents a user set which is extracted to participate in the current round of training; w represents a parameter matrix, w [ J ]]Representing a partial parameter matrix corresponding to the index J; u represents the index of the user embedding parameter, and I represents the index of the online video embedding parameter; u shapekAn index representing user k embedding parameters; i iskAn index of embedded parameters representing the online video in which user k is associated; n denotes the index of the non-embedded parameter.
In the same-class user set, the users extracted to participate in the training of the current round are taken as representative users, and the other users not extracted are taken as subordinate users, so that the updating mode of the model parameters is as follows:
for the embedding parameter representing the user, enabling the representing user to be k E S, and adopting the embedding parameter value updated after local training by the embedding parameter updating rule, namely:
wt+1[Uk]＝wt[Uk]
for the embedding parameters of the subordinate users, because the users in the same user group have certain similarity after clustering operation, the embedding parameters of other subordinate users are related to the updating of the embedding parameters of the representative user extracted to participate in training in the user group, so that the representative user k belongs to S, and the subordinate user C belongs to CkS, the sum of the variance of the user embedding parameters of all the representative users in the user group is delta Uc]＝Σk∈S(wt[Uk]-wt-1[Uk]) The total number of representative users of the user group is | CkAnd n S, γ is the discount rate, which decreases with the increase of the training round, and the dependent user embedding parameter updating rule adopts the original embedding parameter plus the mean value of the variation of the user embedding parameters of all the representative users of the user group, which decreases with the increase of the training round, that is:
for the embedded parameters of the video information, the video request is related to the user, the embedded parameters of the video directly related to the representative user can be correspondingly updated, the representative user k belongs to S, and the influence factor of the user k on all the related videos is
Determined by the sum of the variation before and after the embedded parameter is updated, and each video i belongs to the Uk∈SIkThen, the video embedding parameter updating rule adopts the influence factor of the representative user k on the video i multiplied by the video information embedding parameter of the corresponding representative user and divided by the sum of the influence factors of each representative user on the video, that is:
for the non-embedded parameter part, adopting an updating rule in the traditional federal learning to enable the representative user k to belong to S, nσ＝∑k∈S nkThe update rule of the non-embedded parameters is the weighted average of the whole of all the representative users, i.e. the total number of data trained for all the representative users
According to the federally-learned collaborative online video edge caching method in the embodiment, the method for performing collaborative caching decision by a plurality of edge nodes is as follows:
a1 obtaining a prediction list of a plurality of user online video requests in the coverage area range of each edge node;
a2, calculating the caching value of the online video according to the occurrence frequency of different online videos in the prediction lists of a plurality of users and the participation of the users;
a3 according to the online video caching value, the edge servers of each edge node cooperate to cache until there is no redundant storage space.
Referring to fig. 3, according to the collaborative online video edge caching method based on federal learning of the present embodiment, a plurality of edge nodes perform collaborative service response on online video requests of users.
According to the collaborative online video edge caching method based on the federal learning in the embodiment, the collaborative service response comprises the following steps: when each edge node receives a user online video request in the coverage area of the edge node, the edge node judges whether the online video exists in a cache list of the current node or not: if the local service exists, the edge node carries out local service response, otherwise, the edge node caches the neighbor nodes; and judging whether the online video exists in a cache list of the neighbor node: if the neighbor node exists, the neighbor node performs service response, otherwise, the central server performs service response.
Some steps in the embodiments of the present invention may be implemented by software, and the corresponding software program may be stored in a readable storage medium, such as an optical disc or a hard disk.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (6)
1. A collaborative online video edge caching method based on federal learning, wherein a participating object comprises a plurality of users, a plurality of edge nodes and a central server, the plurality of edge nodes provide services for the plurality of users which can move in different geographical positions in the coverage area of the edge nodes through a wireless cellular network, and are connected to the central server through a backhaul link, and each edge node is provided with an edge server with computing capability and storage capability, the collaborative online video edge caching method is characterized in that: the method comprises the following steps:
step one, establishing a network model according to the plurality of users, the plurality of edge nodes and the central server of the participating objects;
step two, establishing a prediction model by adopting an improved federal learning method, training the prediction model to obtain a user online video request prediction model, and predicting the user online video request; the steps of establishing the prediction model for the online video request of the user by adopting the improved federated learning method are as follows:
s1, distributing a prediction model training instruction to the users by the central server, wherein the prediction model training instruction comprises a method for processing data of request data of the users and selection of the prediction model;
s2, the plurality of users carry out embedding processing on user information and video information in local request data of the users and send the embedded user information to the central server;
s3, clustering the embedded user information by the central server to form a p-type user set; extracting m users from a p-type user set to participate in the training of the prediction model, ensuring that the number of the users randomly extracted from each type of user set is equal, and instructing the m users to perform local training of the prediction model;
s4, the central server initializes the global model parameters of the prediction model and distributes the global model parameters to the extracted m users, each user utilizes local request data to carry out local training on the prediction model, and local model parameters obtained after training are sent to the central server;
s5, the central server aggregates local model parameters of the m users after local training to obtain updated global model parameters, and sends the updated global model parameters to the multiple users to update the model parameters in a new round;
s6, repeating the steps of S2-S5 until the prediction model is converged to obtain a prediction model of the user online video request, and performing request prediction to obtain a prediction list of the user online video request;
and automatically updating the trained prediction model; the automatic updating method comprises the following steps: the plurality of edge nodes record the service response condition of the online video request of the user, calculate the hit rate of the edge cache and the hit rate of the cooperative cache, compare the calculated result with a set threshold, and if the calculated result is lower than the threshold, train the model again to realize automatic updating;
step three, obtaining a user online video request prediction list according to the user online video request prediction model, analyzing the user online video request prediction list in the area coverage range by the edge nodes, and performing edge caching by adopting a collaborative caching decision; the method for the plurality of edge nodes to perform the collaborative cache decision includes:
a1 obtaining a prediction list of user online video requests in the coverage area range of each edge node;
a2, calculating the caching value of the online video according to the occurrence frequency of different online videos in the prediction lists of the users and the participation of the users;
a3 according to the online video caching value, the edge servers of the edge nodes cooperate to cache until there is no redundant storage space.
2. The collaborative online video edge caching method based on federated learning according to claim 1, wherein when the prediction model is updated with parameters, an improved federated learning parameter updating method is used to update the parameters of the prediction model of the user.
3. The collaborative online video edge caching method based on federated learning according to claim 2, characterized in that in the improved federated learning parameter updating method, clustered similar users are collected, users extracted to participate in model training are taken as representative users, and other users not extracted are taken as dependent users;
and updating the parameters of the prediction model of the subordinate user when a next round of model training is needed after the representative user completes local training and the central server completes global parameter updating of the prediction model by utilizing the similarity of the clustered similar users.
4. The collaborative online video edge caching method based on federated learning according to claim 3, wherein the parameters to be updated in the user prediction model are divided into user information embedding parameters, video information embedding parameters and non-embedding parameters, and the parameters are updated by adopting the following rules:
aiming at the user information embedding parameters, the representative user adopts the embedding parameter values updated after local training, and the subordinate user adopts the embedding parameter values of the subordinate user in the previous round and the average value of the variation of the user information embedding parameters of all the representative users in the same user set attenuated along with the increase of the round;
for the video information embedding parameter, the embedding parameter of the video information directly related to the representative user is updated to be the influence factor of the representative user on the video, multiplied by the video information embedding parameter of the corresponding representative user and divided by the sum of the influence factor of each representative user on the video;
aiming at non-embedded parameters, a traditional parameter updating method of federal learning is adopted.
5. The collaborative online video edge caching method based on federated learning of claim 1, wherein the plurality of edge nodes collaboratively serve responses to online video requests of the plurality of users.
6. The collaborative online video edge caching method based on federated learning according to claim 5, wherein the collaborative service response includes: when each edge node receives a user online video request in the coverage area of the edge node, the edge node judges whether the online video requested by the user exists in a cache list of the current node or not: if the local service exists, the edge node carries out local service response, otherwise, the edge node acquires a cache list of the neighbor node; and judging whether the online video exists in a cache list of the neighbor node: if the neighbor node exists, the neighbor node performs service response, otherwise, the central server performs service response.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110521197.3A CN113315978B (en) | 2021-05-13 | 2021-05-13 | Collaborative online video edge caching method based on federal learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110521197.3A CN113315978B (en) | 2021-05-13 | 2021-05-13 | Collaborative online video edge caching method based on federal learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113315978A CN113315978A (en) | 2021-08-27 |
| CN113315978B true CN113315978B (en) | 2022-03-15 |
Family
ID=77373187
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110521197.3A Active CN113315978B (en) | 2021-05-13 | 2021-05-13 | Collaborative online video edge caching method based on federal learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113315978B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113965937B (en) * | 2021-10-27 | 2024-02-13 | 东南大学 | Content popularity prediction method based on clustered federal learning in fog wireless access network |
| CN114302236B (en) * | 2021-12-22 | 2024-04-26 | 江苏号百科技有限公司 | Video distribution method and system based on federal learning |
| CN114584801B (en) * | 2022-01-13 | 2022-12-09 | 北京理工大学 | Video resource caching method based on graph neural network recommendation algorithm |
| CN114595632A (en) * | 2022-03-07 | 2022-06-07 | 北京工业大学 | Mobile edge cache optimization method based on federal learning |
| CN115696296B (en) | 2022-10-25 | 2023-07-07 | 山东大学 | Active edge caching method based on community discovery and weighted federation learning |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112565331A (en) * | 2020-11-02 | 2021-03-26 | 中山大学 | Edge calculation-based end-edge collaborative federated learning optimization method |
| CN112752308A (en) * | 2020-12-31 | 2021-05-04 | 厦门越人健康技术研发有限公司 | Mobile prediction wireless edge caching method based on deep reinforcement learning |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107909108B (en) * | 2017-11-15 | 2021-06-11 | 东南大学 | Edge cache system and method based on content popularity prediction |
| WO2020115273A1 (en) * | 2018-12-07 | 2020-06-11 | Telefonaktiebolaget Lm Ericsson (Publ) | Predicting network communication performance using federated learning |
| US11323514B2 (en) * | 2019-04-30 | 2022-05-03 | EMC IP Holding Company LLC | Data tiering for edge computers, hubs and central systems |
| CN110263921B (en) * | 2019-06-28 | 2021-06-04 | 深圳前海微众银行股份有限公司 | Method and device for training federated learning model |
| CN110929880A (en) * | 2019-11-12 | 2020-03-27 | 深圳前海微众银行股份有限公司 | Method and device for federated learning and computer readable storage medium |
| CN111277860B (en) * | 2020-01-23 | 2021-04-06 | 北京邮电大学 | Method, device and equipment for caching video in mobile edge network and readable medium |
| CN111865826B (en) * | 2020-07-02 | 2022-01-04 | 大连理工大学 | Active content caching method based on federal learning |
| CN111935784B (en) * | 2020-08-12 | 2022-04-22 | 重庆邮电大学 | Content caching method based on federal learning in fog computing network |
| CN112118295A (en) * | 2020-08-21 | 2020-12-22 | 深圳大学 | File caching method and device, edge node and computer readable storage medium |
-
2021
- 2021-05-13 CN CN202110521197.3A patent/CN113315978B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112565331A (en) * | 2020-11-02 | 2021-03-26 | 中山大学 | Edge calculation-based end-edge collaborative federated learning optimization method |
| CN112752308A (en) * | 2020-12-31 | 2021-05-04 | 厦门越人健康技术研发有限公司 | Mobile prediction wireless edge caching method based on deep reinforcement learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113315978A (en) | 2021-08-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113315978B (en) | Collaborative online video edge caching method based on federal learning | |
| CN111835827B (en) | Internet of things edge computing task unloading method and system | |
| CN111901392B (en) | Mobile edge computing-oriented content deployment and distribution method and system | |
| Wang et al. | A novel reputation-aware client selection scheme for federated learning within mobile environments | |
| CN114357455B (en) | Trust method based on multidimensional attribute trust evaluation | |
| CN113255004A (en) | Safe and efficient federal learning content caching method | |
| CN115633380B (en) | Multi-edge service cache scheduling method and system considering dynamic topology | |
| CN115941790A (en) | Edge collaborative content caching method, device, equipment and storage medium | |
| CN116346837A (en) | Internet of things edge collaborative caching method based on deep reinforcement learning | |
| CN114281718A (en) | Industrial Internet edge service cache decision method and system | |
| Li et al. | Human in the loop: distributed deep model for mobile crowdsensing | |
| CN114374949B (en) | Information freshness optimization-based power control mechanism in Internet of vehicles | |
| CN115314944A (en) | Internet of vehicles cooperative caching method based on mobile vehicle social relation perception | |
| CN111770152B (en) | Edge data management method, medium, edge server and system | |
| CN111885551B (en) | Selection and allocation mechanism of high-influence users in multi-mobile social network based on edge cloud collaborative mode | |
| CN117459112A (en) | Mobile edge caching method and equipment in LEO satellite network based on graph rolling network | |
| CN115587266A (en) | Air-space-ground integrated internet intelligent edge caching method | |
| KR102235622B1 (en) | Method and Apparatus for Cooperative Edge Caching in IoT Environment | |
| CN115129888A (en) | Active content caching method based on network edge knowledge graph | |
| CN115189910A (en) | Network digital twin-based deliberate attack survivability evaluation method | |
| Khodaparas et al. | An Intelligent Caching Scheme Considering the Spatio-Temporal Characteristics of Data in Internet of Vehicles | |
| CN114519150A (en) | Mutual trust evaluation method and system for Internet of things network | |
| CN107609039B (en) | CDN-based space distribution method for hot films at near-end cache server | |
| CN111901394A (en) | Method and system for caching moving edge by jointly considering user preference and activity degree | |
| CN114285748B (en) | Reputation evaluation method and system based on Internet of things |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |