CN113971461A - Distributed federal learning method and system for unmanned aerial vehicle ad hoc network - Google Patents
Distributed federal learning method and system for unmanned aerial vehicle ad hoc network Download PDFInfo
- Publication number
- CN113971461A CN113971461A CN202111251156.3A CN202111251156A CN113971461A CN 113971461 A CN113971461 A CN 113971461A CN 202111251156 A CN202111251156 A CN 202111251156A CN 113971461 A CN113971461 A CN 113971461A
- Authority
- CN
- China
- Prior art keywords
- unmanned aerial
- aerial vehicle
- model
- training
- round
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Abstract
The invention discloses an unmanned aerial vehicle ad hoc network distributed federal learning method, which comprises the following steps: in each round of training process, each unmanned aerial vehicle receives local model parameters sent by a plurality of single-hop neighbor nodes, and the received local models of the plurality of single-hop neighbor nodes are aggregated with the local model of the unmanned aerial vehicle; updating the model on the basis of the obtained aggregation model to generate a new round of local model; broadcasting the updated new round of local model parameters to each neighbor node; before each round of training is started, the unmanned aerial vehicle ad hoc network structure is reconstructed to update the single-hop neighbor node list of each unmanned aerial vehicle, and then the round of training is started. The method greatly improves the robustness of the federal learning of the unmanned aerial vehicle ad hoc network, and simultaneously can improve the flexibility and agility of the federal learning in the unmanned aerial vehicle network, so that the federal learning can be better adapted to the unmanned aerial vehicle network with high and dynamic network topology.
Description
Technical Field
The invention relates to the technical field of unmanned aerial vehicle cluster edge intelligence, in particular to an unmanned aerial vehicle ad hoc network distributed federal learning method and system.
Background
With the development of drone technology, drones are expected to play a key role in numerous applications in the next generation of wireless networks, from cargo delivery, target surveillance to telecommunications applications in the civilian and military fields. On one hand, due to its flexibility, line-of-sight connectivity and 3D mobility, the drone can serve as a flying base station to provide communication/computation/caching services in future wireless networks, which make up for the deficiencies of traditional infrastructure-based networks; on the other hand, the unmanned aerial vehicle can also be used as a flight user, and supports emerging applications such as remote sensing, goods delivery, target identification and tracking and the like.
While machine learning possesses the ability to enable drone network intelligence, traditional machine learning approaches are cloud-centric, i.e., all data needs to be transmitted to and processed in a cloud data center, which may not be well suited for drone networks. First, due to privacy concerns, the data generated by each drone may not be accessible because it may contain some information that is sensitive to, for example, the identity and location of the drone; secondly, for some real-time drone applications, such as automatic drone monitoring and target tracking, the delay from sending raw data to receiving a well-trained model is unacceptable; finally, transmitting huge raw data such as images and videos to the cloud consumes a lot of bandwidth and energy, which is unacceptable for drone networks with limited bandwidth and energy supply. Federal learning is a promising distributed machine learning, and can enable a plurality of devices to cooperatively train a machine learning model under the condition of not sending original data, so that the privacy of the devices is protected, experience delay is improved, and bandwidth and energy burden are reduced.
However, there remain some challenges in applying the conventional federal learning approach based on centralized federal learning to drone swarm networks. Different from a relatively reliable parameter server in the ground federal learning, the centralized federal learning inevitably faces a single point of failure when applied to an unmanned aerial vehicle network, and when the unmanned aerial vehicle of the parameter server receives an attack or a battery failure and cannot normally work, the training of the whole unmanned aerial vehicle cluster is forced to be interrupted. For example, in view of a battlefield denial environment, deploying conventional centralized-based federal learning on a cluster of drones for patrol missions faces many challenges, the two largest of which are as follows: 1) when the centralized federal learning is applied to an unmanned aerial vehicle network, a single point of failure is inevitably faced, and when the parameter server unmanned aerial vehicle receives an attack or a battery failure and cannot normally work, the training of the whole unmanned aerial vehicle cluster is forced to be interrupted; 2) due to the high dynamic characteristics of the unmanned aerial vehicles, the network topology changes due to the dynamic connection and disconnection of the unmanned aerial vehicles with certain training nodes on the basis of centralized federal learning network topology, and at the moment, the whole federal learning needs to be recombined.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides the distributed federal learning method and the distributed federal learning system for the unmanned aerial vehicle ad hoc network, which are different from the conventional federal learning method based on centralization, wherein model parameters of training nodes need to be converged to a central parameter server for model aggregation. By the method, robustness of the self-organizing network federal learning of the unmanned aerial vehicle is greatly improved, flexibility and agility of the federal learning in an unmanned aerial vehicle network can be improved, and the federal learning can be better adapted to the unmanned aerial vehicle network with high and dynamic network topology.
In order to achieve the purpose, the invention adopts the following technical scheme:
an unmanned aerial vehicle ad hoc network distributed federal learning method, the unmanned aerial vehicle cluster that the said unmanned aerial vehicle ad hoc network corresponds to is made up of many unmanned aerial vehicles with local data set and model training ability;
the learning method includes the steps of:
in each round of training process, each unmanned aerial vehicle receives local model parameters sent by a plurality of single-hop neighbor nodes, and the received local models of the plurality of single-hop neighbor nodes are aggregated with the local model of the unmanned aerial vehicle; updating the model on the basis of the obtained aggregation model to generate a new round of local model; broadcasting the updated new round of local model parameters to each neighbor node;
before each round of training is started, the unmanned aerial vehicle ad hoc network structure is reconstructed to update the single-hop neighbor node list of each unmanned aerial vehicle, and then the round of training is started.
Further, before each round of training begins, reconstructing the drone ad hoc network structure to update the single-hop neighbor node list of each drone includes:
when any neighbor node is attacked, failed or departed, deleting the neighbor node from the single-hop neighbor node list; and when a new node is added, generating a single-hop neighbor node list of the node according to the network characteristics of the node and updating the single-hop neighbor node list of the corresponding unmanned aerial vehicle.
Further, the learning process of any one drone i includes the following steps:
s1, in the training of the t round, the unmanned aerial vehicle i firstly receives the local model parameters W sent by K single-hop neighbor nodesi+1，t，Wi+2，t...Wi+K，t(ii) a N, N is the total number of drones in the drone cluster; t is a positive integer greater than or equal to 1;
s2, carrying out model aggregation on the received local models and the local model of the node:
in the formula (I), the compound is shown in the specification,
representing the parameters of the aggregation model obtained after model aggregation of the unmanned aerial vehicle i during the t-th round of training, Di+kRepresenting the local dataset size for the i + k-th drone;
s3, updating the model on the basis of the obtained aggregation model to generate a new round of local model:
in the formula, Wi，t+1Representing a new round of local model parameters updated by the drone i with the aggregated model parameters,
representing pairs of cross-entropy loss functions
λ represents the learning rate;
s4, broadcasting the updated new round of local model parameters to each neighbor node, so as to transmit the local model from the unmanned aerial vehicle i to each neighbor node, and enabling the neighbor nodes to perform model aggregation in the next round of training.
Further, the local data sets of each drone are independent of each other.
Further, the learning method further includes the steps of:
and evaluating the learning effect by respectively adopting the training loss value of each unmanned aerial vehicle individual, the average training loss value of the unmanned aerial vehicle ad hoc network and the training delay of the unmanned aerial vehicle ad hoc network.
Based on the distributed federal learning method, the invention also provides an unmanned aerial vehicle ad hoc network distributed federal learning system, wherein an unmanned aerial vehicle cluster corresponding to the unmanned aerial vehicle ad hoc network is composed of a plurality of unmanned aerial vehicles with local data sets and model training capability;
the distributed federal learning system comprises a learning device installed on each unmanned aerial vehicle; the learning device includes:
the model acquisition module is used for receiving local model parameters sent by a plurality of single-hop neighbor nodes in each round of training process;
the aggregation module is used for aggregating the received local models of the multiple single-hop neighbor nodes with the local model of the unmanned aerial vehicle;
the model updating module is used for updating the model on the basis of the obtained aggregation model to generate a new round of local model;
the broadcasting module is used for broadcasting the updated new round of local model parameters to each neighbor node;
and the node list maintenance module is used for reconstructing the ad hoc network structure of the unmanned aerial vehicle to update the single-hop neighbor node list of the current unmanned aerial vehicle before each round of training is started.
The invention has the beneficial effects that:
(1) the distributed federal learning method of the unmanned aerial vehicle ad hoc network is different from the conventional centralized federal learning method which needs to gather the model parameters of the training nodes to the central parameter server for model aggregation. By the method, robustness of the self-organizing network federal learning of the unmanned aerial vehicle is greatly improved, flexibility and agility of the federal learning in an unmanned aerial vehicle network can be improved, and the federal learning can be better adapted to the unmanned aerial vehicle network with high and dynamic network topology.
(2) Practice proves that because the conventional unmanned aerial vehicle network federal learning needs to broadcast the aggregated global model in each round of communication, the training delay of distributed unmanned aerial vehicle network federal learning is always smaller than that of the conventional unmanned aerial vehicle network federal learning, and the final reduced delay is about 101.72 milliseconds after 60 rounds of training.
Drawings
Fig. 1 is a flowchart of a distributed federal learning method for an ad hoc network of an unmanned aerial vehicle and a corresponding architecture diagram.
Fig. 2(a) is a scene diagram of a simulation experiment of conventional unmanned aerial vehicle network federal learning.
Fig. 2(b) is a scene diagram of a simulation experiment of distributed drone network federal learning according to an embodiment of the present invention.
Fig. 3 is a simulation result diagram of average training loss of conventional unmanned aerial vehicle network federal learning and distributed unmanned aerial vehicle network federal learning.
Fig. 4 is a training loss simulation result diagram of each individual unmanned aerial vehicle under the conventional unmanned aerial vehicle network federal learning and the distributed unmanned aerial vehicle network federal learning.
Fig. 5 is a simulation result diagram of training time delay of conventional unmanned aerial vehicle network federal learning and distributed unmanned aerial vehicle network federal learning.
Detailed Description
The present invention will now be described in further detail with reference to the accompanying drawings.
It should be noted that the terms "upper", "lower", "left", "right", "front", "back", etc. used in the present invention are for clarity of description only, and are not intended to limit the scope of the present invention, and the relative relationship between the terms and the terms is not limited by the technical contents of the essential changes.
The embodiment provides an unmanned aerial vehicle ad hoc network distributed federal learning method, wherein an unmanned aerial vehicle cluster is composed of a plurality of unmanned aerial vehicles which are provided with local data sets and have certain model training capacity, the unmanned aerial vehicle cluster carries out battlefield patrol tasks according to a preset air route, federal learning is carried out on the unmanned aerial vehicle cluster in the patrol process, and a learning model is trained to carry out sensing recognition on enemy combat units. Fig. 1 is a flowchart of a distributed federal learning method for an ad hoc network of an unmanned aerial vehicle and a corresponding architecture diagram. The unmanned aerial vehicle cluster is composed of a plurality of unmanned aerial vehicles with local data sets and certain model training capacity, each unmanned aerial vehicle can be used as a parameter server of a neighbor node in a single-hop range, and received local models of the neighbor node and local models of the neighbor node are aggregated.
Referring to fig. 1, taking an unmanned aerial vehicle i as an example, a process of each unmanned aerial vehicle node executing a round of global training specifically includes the following steps:
(1) in the training of the t-th round, an unmanned aerial vehicle i firstly receives local model parameters W sent by K single-hop neighbor nodesi+1，t，Wi+2，t…Wi+K，t。
(2) Carrying out model aggregation on the received local models and the local model of the node:
(3) Updating the model on the basis of the obtained aggregation model to generate a new round of local model:
Wi，t+1representing a new round of local model parameters updated by the drone i with the aggregated model parameters,
representing pairs of cross-entropy loss functions
λ represents the learning rate.
(4) And broadcasting the updated local model parameters of the new round to each neighbor node so as to transmit the local model from the unmanned aerial vehicle i to each neighbor node, so that the neighbor nodes can perform model aggregation in the next round of training. It should be understood that other nodes also have functions similar to the unmanned aerial vehicle i, and it is equivalent to that each unmanned aerial vehicle is not only a training node but also a parameter server, so as to implement a distributed architecture in a true sense.
Based on the distributed federal learning method, the embodiment of the invention also provides an unmanned aerial vehicle ad hoc network distributed federal learning system, wherein an unmanned aerial vehicle cluster corresponding to the unmanned aerial vehicle ad hoc network is composed of a plurality of unmanned aerial vehicles with local data sets and model training capability.
The distributed federal learning system comprises a learning device installed on each unmanned aerial vehicle; the learning device comprises a model acquisition module, an aggregation module, a model updating module, a broadcasting module and a node list maintenance module.
And the model acquisition module is used for receiving local model parameters sent by a plurality of single-hop neighbor nodes in each round of training process.
And the aggregation module is used for aggregating the received local models of the multiple single-hop neighbor nodes with the local model of the unmanned aerial vehicle.
And the model updating module is used for updating the model on the basis of the obtained aggregation model to generate a new round of local model.
And the broadcasting module is used for broadcasting the updated new round of local model parameters to each neighbor node.
And the node list maintenance module is used for reconstructing the ad hoc network structure of the unmanned aerial vehicle to update the single-hop neighbor node list of the current unmanned aerial vehicle before each round of training is started.
Fig. 2(a) and 2(b) are simulation experiment scene diagrams of conventional unmanned aerial vehicle network federal learning and distributed unmanned aerial vehicle network federal learning respectively. In a conventional unmanned aerial vehicle network federal learning scene, a first unmanned aerial vehicle serves as a parameter server to receive local models of other five training node unmanned aerial vehicles for model aggregation; in distributed unmanned aerial vehicle network federal learning, if an unmanned aerial vehicle I cannot normally work due to attack, other five unmanned aerial vehicles are used as parameter servers of neighbor nodes in a single-hop range according to the distributed federal learning method provided by the invention, each unmanned aerial vehicle is a training node, and the received local model of the neighbor node and the local model of the neighboring node are aggregated, and model updating is carried out on the basis. Fig. 3 is a graph of simulation results of average training loss of conventional network federal learning and distributed network federal learning of unmanned aerial vehicles, and it can be seen that although the average loss value of distributed network federal learning of unmanned aerial vehicles is always higher than that of conventional network federal learning of unmanned aerial vehicles during training, when the training effect converges, the final difference between the average loss values is only 0.0156. This means that distributed drone network federal learning is almost as effective as conventional drone network federal learning without the help of a central parameter server.
Fig. 4 is a training loss simulation result diagram of each individual unmanned aerial vehicle under the conventional unmanned aerial vehicle network federal learning and the distributed unmanned aerial vehicle network federal learning. After 60 rounds of training, the loss difference between the distributed network federal learning of drones and the conventional network federal learning of drones is very small for each drone, and is only not more than 0.0229 (as can be seen from fig. 3, the loss value before untraining is close to 2). On the other hand, the individual loss of each drone may also be different, because in distributed network federal learning of drones, the local model is not updated on the global federal learning model and the data set carried by each drone is different.
Fig. 5 is a result chart of a training experiment of the conventional network federal learning and the distributed network federal learning of the unmanned aerial vehicle, and it can be seen from the result chart that the training delay of the distributed network federal learning of the unmanned aerial vehicle is always smaller than that of the conventional network federal learning of the unmanned aerial vehicle, wherein the final reduction delay is about 101.72 milliseconds, because the conventional network federal learning of the unmanned aerial vehicle needs to broadcast the aggregated global model in each round of communication. Meanwhile, the training delay curves of the two schemes are linear with the number of communication rounds, mainly because the network topology and the resource allocation of each round of communication are the same, so that the training delay of each round is the same
The distributed unmanned aerial vehicle network federal learning method provided by the invention can be well adapted to an unmanned aerial vehicle network with high dynamic network topology, and the distributed unmanned aerial vehicle network federal learning method brings the following two main advantages when improving the intelligence of equipment in the unmanned aerial vehicle network: on the one hand, the distributed unmanned aerial vehicle network federal learning method can perform distributed machine learning with high robustness on an unmanned aerial vehicle network. Specifically, because there is no central node coordinating the learning process in the distributed unmanned aerial vehicle network federal learning method, if any unmanned aerial vehicle or air-to-air link is unavailable, the federal learning will not be terminated; on the other hand, the distributed unmanned aerial vehicle network federal learning method provides high flexibility and agility. Federal learning also does not require reassembly regardless of how the network topology changes due to dynamic connections and disconnections. In short, the distributed drone network federal learning approach is novel in that it proposes a federal learning framework that fully adapts to the characteristics of the drone network.
The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above-mentioned embodiments, and all technical solutions belonging to the idea of the present invention belong to the protection scope of the present invention. It should be noted that modifications and embellishments within the scope of the invention may be made by those skilled in the art without departing from the principle of the invention.
Claims (6)
1. An unmanned aerial vehicle ad hoc network distributed federal learning method is characterized in that an unmanned aerial vehicle cluster corresponding to the unmanned aerial vehicle ad hoc network is composed of a plurality of unmanned aerial vehicles which are provided with local data sets and have model training capability;
the learning method includes the steps of:
in each round of training process, each unmanned aerial vehicle receives local model parameters sent by a plurality of single-hop neighbor nodes, and the received local models of the plurality of single-hop neighbor nodes are aggregated with the local model of the unmanned aerial vehicle; updating the model on the basis of the obtained aggregation model to generate a new round of local model; broadcasting the updated new round of local model parameters to each neighbor node;
before each round of training is started, the unmanned aerial vehicle ad hoc network structure is reconstructed to update the single-hop neighbor node list of each unmanned aerial vehicle, and then the round of training is started.
2. The distributed federated learning method of ad hoc networks of unmanned aerial vehicles of claim 1, wherein reconstructing the ad hoc network structure of unmanned aerial vehicles to update the list of single-hop neighbor nodes of each unmanned aerial vehicle before each round of training begins comprises:
when any neighbor node is attacked, failed or departed, deleting the neighbor node from the single-hop neighbor node list; and when a new node is added, generating a single-hop neighbor node list of the node according to the network characteristics of the node and updating the single-hop neighbor node list of the corresponding unmanned aerial vehicle.
3. The distributed federated learning method for ad hoc networks of unmanned aerial vehicles according to claim 1, wherein the learning process for any unmanned aerial vehicle i comprises the following steps:
s1, in the training of the t round, the unmanned aerial vehicle i firstly receives the local model parameters W sent by K single-hop neighbor nodesi+1，t，Wi+2，t...Wi+K，t(ii) a N, N is the total number of drones in the drone cluster; t is a positive integer greater than or equal to 1;
s2, carrying out model aggregation on the received local models and the local model of the node:
in the formula (I), the compound is shown in the specification,
representing the parameters of the aggregation model obtained after model aggregation of the unmanned aerial vehicle i during the t-th round of training, Di+kRepresenting the local dataset size for the i + k-th drone;
s3, updating the model on the basis of the obtained aggregation model to generate a new round of local model:
in the formula, Wi，t+1Representing a new round of local model parameters updated by the drone i with the aggregated model parameters,
representing pairs of cross-entropy loss functions
λ represents the learning rate;
s4, broadcasting the updated new round of local model parameters to each neighbor node, so as to transmit the local model from the unmanned aerial vehicle i to each neighbor node, and enabling the neighbor nodes to perform model aggregation in the next round of training.
4. The distributed federated learning method of ad hoc networking of unmanned aerial vehicles of claim 3, wherein the local data sets of each unmanned aerial vehicle are independent of each other.
5. The distributed federated learning method of ad hoc networks of unmanned aerial vehicles according to claim 1, wherein the learning method further comprises the steps of:
and evaluating the learning effect by respectively adopting the training loss value of each unmanned aerial vehicle individual, the average training loss value of the unmanned aerial vehicle ad hoc network and the training delay of the unmanned aerial vehicle ad hoc network.
6. An unmanned aerial vehicle ad hoc network distributed federal learning system based on the distributed federal learning method of any one of claims 1 to 5, wherein the unmanned aerial vehicle cluster corresponding to the unmanned aerial vehicle ad hoc network is composed of a plurality of unmanned aerial vehicles with local data sets and model training capability;
the distributed federal learning system comprises a learning device installed on each unmanned aerial vehicle; the learning device includes:
the model acquisition module is used for receiving local model parameters sent by a plurality of single-hop neighbor nodes in each round of training process;
the aggregation module is used for aggregating the received local models of the multiple single-hop neighbor nodes with the local model of the unmanned aerial vehicle;
the model updating module is used for updating the model on the basis of the obtained aggregation model to generate a new round of local model;
the broadcasting module is used for broadcasting the updated new round of local model parameters to each neighbor node;
and the node list maintenance module is used for reconstructing the ad hoc network structure of the unmanned aerial vehicle to update the single-hop neighbor node list of the current unmanned aerial vehicle before each round of training is started.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111251156.3A CN113971461A (en) | 2021-10-26 | 2021-10-26 | Distributed federal learning method and system for unmanned aerial vehicle ad hoc network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111251156.3A CN113971461A (en) | 2021-10-26 | 2021-10-26 | Distributed federal learning method and system for unmanned aerial vehicle ad hoc network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113971461A true CN113971461A (en) | 2022-01-25 |
Family
ID=79588497
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111251156.3A Pending CN113971461A (en) | 2021-10-26 | 2021-10-26 | Distributed federal learning method and system for unmanned aerial vehicle ad hoc network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113971461A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114530028A (en) * | 2022-02-14 | 2022-05-24 | 大连理工大学 | Campus student intelligent bracelet monitoring system and method based on LoRa communication and federal learning |
| CN115173924A (en) * | 2022-07-06 | 2022-10-11 | 多彩贵州印象网络传媒股份有限公司 | Unmanned aerial vehicle unmanned inspection management system based on cloud edge cooperation technology |
| CN116582871A (en) * | 2023-07-07 | 2023-08-11 | 南京邮电大学 | Unmanned aerial vehicle cluster federal learning model optimization method based on topology optimization |
| CN116778363A (en) * | 2023-06-25 | 2023-09-19 | 河海大学 | Low-traffic reservoir area water environment risk identification method based on federal learning |
| CN117332878A (en) * | 2023-10-31 | 2024-01-02 | 慧之安信息技术股份有限公司 | Model training method and system based on ad hoc network system |
| CN116778363B (en) * | 2023-06-25 | 2024-04-30 | 河海大学 | Low-traffic reservoir area water environment risk identification method based on federal learning |
-
2021
- 2021-10-26 CN CN202111251156.3A patent/CN113971461A/en active Pending
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114530028A (en) * | 2022-02-14 | 2022-05-24 | 大连理工大学 | Campus student intelligent bracelet monitoring system and method based on LoRa communication and federal learning |
| CN115173924A (en) * | 2022-07-06 | 2022-10-11 | 多彩贵州印象网络传媒股份有限公司 | Unmanned aerial vehicle unmanned inspection management system based on cloud edge cooperation technology |
| CN116778363A (en) * | 2023-06-25 | 2023-09-19 | 河海大学 | Low-traffic reservoir area water environment risk identification method based on federal learning |
| CN116778363B (en) * | 2023-06-25 | 2024-04-30 | 河海大学 | Low-traffic reservoir area water environment risk identification method based on federal learning |
| CN116582871A (en) * | 2023-07-07 | 2023-08-11 | 南京邮电大学 | Unmanned aerial vehicle cluster federal learning model optimization method based on topology optimization |
| CN116582871B (en) * | 2023-07-07 | 2023-10-13 | 南京邮电大学 | Unmanned aerial vehicle cluster federal learning model optimization method based on topology optimization |
| CN117332878A (en) * | 2023-10-31 | 2024-01-02 | 慧之安信息技术股份有限公司 | Model training method and system based on ad hoc network system |
| CN117332878B (en) * | 2023-10-31 | 2024-04-16 | 慧之安信息技术股份有限公司 | Model training method and system based on ad hoc network system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113971461A (en) | Distributed federal learning method and system for unmanned aerial vehicle ad hoc network | |
| Wang et al. | Transfer learning promotes 6G wireless communications: Recent advances and future challenges | |
| CN108616302B (en) | Unmanned aerial vehicle multiple coverage model under power control and deployment method | |
| Zhang et al. | Federated learning assisted multi-UAV networks | |
| Pham et al. | Aerial computing: A new computing paradigm, applications, and challenges | |
| Qu et al. | Decentralized federated learning for UAV networks: Architecture, challenges, and opportunities | |
| CN108882201B (en) | Energy efficiency-oriented relay transmission selection method for unmanned aerial vehicle cluster network | |
| CN112737842A (en) | Task safety unloading method based on minimized time delay in air-ground integrated Internet of vehicles | |
| CN113660681B (en) | Multi-agent resource optimization method applied to unmanned aerial vehicle cluster auxiliary transmission | |
| Mohammed et al. | Towards trusted and efficient UAV-based communication | |
| Pham et al. | Aerial access networks for federated learning: Applications and challenges | |
| Qu et al. | Empowering the edge intelligence by air-ground integrated federated learning in 6G networks | |
| Xu et al. | Efficient deployment of multi‐UAV assisted mobile edge computing: A cost and energy perspective | |
| CN114070379A (en) | Unmanned aerial vehicle flight path optimization and resource allocation method based on safety energy efficiency fairness | |
| Xiao et al. | Implementation for UAVs aided edge sensing system in wireless emergency communications | |
| CN116073924B (en) | Anti-interference channel allocation method and system based on Stackelberg game | |
| Xiao et al. | Fully decentralized federated learning based beamforming design for UAV communications | |
| CN110673651A (en) | Robust formation method for unmanned aerial vehicle cluster under limited communication condition | |
| Shabanighazikelayeh et al. | Optimal UAV deployment for rate maximization in IoT networks | |
| Ying et al. | Game theoretical bandwidth allocation in UAV-UGV collaborative disaster relief networks | |
| Wang et al. | UAV-based physical-layer intelligent technologies for 5G-enabled internet of things: A survey | |
| CN115102767B (en) | DDoS active defense system and method based on distributed collaborative learning | |
| Wang et al. | Airborne computing platform based joint offload and deployment optimization algorithm | |
| Yin et al. | Resource allocation for UAV-assisted MIMO-NOMA wireless caching networks | |
| Song et al. | Personalized Federated Deep Reinforcement Learning-based Trajectory Optimization for Multi-UAV Assisted Edge Computing |
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 |