Bin-Jie Hu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Bin-Jie Hu has a total of 78 co-authors. Here are the top co-authors by number of shared publications:
- Xiu Yin Zhang
South China University of Technology
12 shared publications - Quan Xue
South China University of Technology
7 shared publications - Di-Si Wu
School of Electronic and Information Engineering, South China University of Technology, Guangzhou, China
5 shared publications - Peng Fei Hu
School of Electronics and Information Technology (School of Microelectronics), Guangdong Provincial Key Laboratory of Optoelectronic Information Processing Chips and Systems, Sun Yat-sen University, Guangzhou, China
5 shared publications - Yuan Chun Li
School of Electronic and Information Engineering, South China University of Technology, Guangzhou, Guangdong, China
5 shared publications - Enjun Xia
School of Electronic and Information Engineering, South China University of Technology, Guangzhou, China
4 shared publications
Latest Publications
- 19 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 75 (2) , 1899-1911
A Transformer-Based Trust Management System for the Internet of Vehicles Using Blockchain Technology
In the Internet of Vehicles (IoV), the vehicular network serves as an open environment for information exchange, which makes it susceptible to internal attacks and leads to security concerns. Trust among vehicles plays a crucial role in addressing these challenges. However, previous approaches to trust management lack precision in evaluating vehicle trust, which limits their effectiveness in countering internal attacks within the vehicular network. To accurately assess the trust among vehicles, mitigate the spread of malicious messages, and prevent interference from malicious nodes within the vehicular network, this paper proposes TrustFormer, a trust management system based on Transformer. TrustFormer employs a transformer-based message detection model (TrMD) to detect each message in real time, converting it into a change in vehicle trust value. In addition, roadside units (RSUs) utilize blockchain technology to store the global trust values of vehicles, thereby ensuring their authenticity. Simulation results demonstrate that the proposed method outperforms existing approaches in terms of Precision, Recall, and F1-score against Bad Mouth Attack (BMA) and Zigzag Attack (ZA). Even with an increased number of malicious nodes in the network, the proposed method exhibits significant advantages, with less degradation in performance metrics compared to existing methods. Furthermore, blockchain performance simulation experiments indicate that the system latency meets the requirements of vehicular network applications, and throughput tests confirm its scalability.
- 19 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 75 (2) , 1899-1911
A Transformer-Based Trust Management System for the Internet of Vehicles Using Blockchain Technology
In the Internet of Vehicles (IoV), the vehicular network serves as an open environment for information exchange, which makes it susceptible to internal attacks and leads to security concerns. Trust among vehicles plays a crucial role in addressing these challenges. However, previous approaches to trust management lack precision in evaluating vehicle trust, which limits their effectiveness in countering internal attacks within the vehicular network. To accurately assess the trust among vehicles, mitigate the spread of malicious messages, and prevent interference from malicious nodes within the vehicular network, this paper proposes TrustFormer, a trust management system based on Transformer. TrustFormer employs a transformer-based message detection model (TrMD) to detect each message in real time, converting it into a change in vehicle trust value. In addition, roadside units (RSUs) utilize blockchain technology to store the global trust values of vehicles, thereby ensuring their authenticity. Simulation results demonstrate that the proposed method outperforms existing approaches in terms of Precision, Recall, and F1-score against Bad Mouth Attack (BMA) and Zigzag Attack (ZA). Even with an increased number of malicious nodes in the network, the proposed method exhibits significant advantages, with less degradation in performance metrics compared to existing methods. Furthermore, blockchain performance simulation experiments indicate that the system latency meets the requirements of vehicular network applications, and throughput tests confirm its scalability.
- Open Access
- 1 March 2025
- Published by Tsinghua University Press in Journal of Highway and Transportation Research and Development (English Edition)
- Vol. 19 (1) , 13-17
Lightweight framework for misbehavior detection in internet of vehicles
The proliferation of connected vehicles in the Internet of Vehicles (IoV) ecosystem has introduced new security challenges, particularly in the context of internal network attacks. Traditional public key infrastructure (PKI) technologies are no longer sufficient to ensure secure communication within a network that experiences dynamic topology changes and high vehicle density. In response, there is a growing need for a lightweight misbehavior detection framework that offers fast computation and minimal space complexity. This paper presents a novel approach using continuous-time recurrent neural networks for detecting misbehavior in the IoV and assesses their performance against the Vehicular Reference Misbehavior (VeReMi) extension dataset. We compare two recently introduced models—the liquid time-constant (LTC) network and the closed-form continuous-time (CFC) neural network—with the established convolutional neural network-long short-term memory (CNN-LSTM) model. The results indicate that continuous-time neural networks marginally outperform CNN-LSTM on evaluation metrics. Despite LTC and CFC having significantly fewer parameters, making them less complex and more space-efficient than CNN-LSTM, the latter proves to be more time-efficient. Therefore, a careful balance between runtime cost and space complexity must be considered when deploying lightweight neural networks in practical applications.
- 1 March 2025
- Published by Tsinghua University Press in Journal of Highway and Transportation Research and Development (English Edition)
- Vol. 19 (1) , 13-17
Open AccessLightweight framework for misbehavior detection in internet of vehicles
The proliferation of connected vehicles in the Internet of Vehicles (IoV) ecosystem has introduced new security challenges, particularly in the context of internal network attacks. Traditional public key infrastructure (PKI) technologies are no longer sufficient to ensure secure communication within a network that experiences dynamic topology changes and high vehicle density. In response, there is a growing need for a lightweight misbehavior detection framework that offers fast computation and minimal space complexity. This paper presents a novel approach using continuous-time recurrent neural networks for detecting misbehavior in the IoV and assesses their performance against the Vehicular Reference Misbehavior (VeReMi) extension dataset. We compare two recently introduced models—the liquid time-constant (LTC) network and the closed-form continuous-time (CFC) neural network—with the established convolutional neural network-long short-term memory (CNN-LSTM) model. The results indicate that continuous-time neural networks marginally outperform CNN-LSTM on evaluation metrics. Despite LTC and CFC having significantly fewer parameters, making them less complex and more space-efficient than CNN-LSTM, the latter proves to be more time-efficient. Therefore, a careful balance between runtime cost and space complexity must be considered when deploying lightweight neural networks in practical applications.
- 11 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 74 (6) , 9656-9670
Cuckoo Search-Enabled Task Scheduling and Cache Updating in Vehicular Edge-Fog Computing
With the emergence of computation-intensive vehicle applications, vehicle edge-fog computing (VEFC) is playing an increasingly important role in intelligent transportation systems. In this article, we study the task scheduling and cache updating (TSCU) problem in VEFC, where vehicles can choose three modes to process their tasks: local computing, vehicle-to-vehicle (V2V) offloading, and vehicle-to-infrastructure (V2I) offloading. Service providers can dynamically update their caches to improve the system offloading efficiency. To maximize the offloading efficiency, we formulate the TSCU problem as a mixed-integer non-linear programming (MINLP) problem and develop an improved discrete cuckoo search algorithm to solve the optimization problem. Our proposed algorithm explores the global environment via Levy flight. On the other hand, the algorithm convergence rate is accelerated by generating the initial population based on a greedy algorithm and its variants and updating the discard probability. Simulation results show that our proposal has achieved an up to 12% improvement in offloading efficiency compared to the benchmark schemes.
- 11 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 74 (6) , 9656-9670
Cuckoo Search-Enabled Task Scheduling and Cache Updating in Vehicular Edge-Fog Computing
With the emergence of computation-intensive vehicle applications, vehicle edge-fog computing (VEFC) is playing an increasingly important role in intelligent transportation systems. In this article, we study the task scheduling and cache updating (TSCU) problem in VEFC, where vehicles can choose three modes to process their tasks: local computing, vehicle-to-vehicle (V2V) offloading, and vehicle-to-infrastructure (V2I) offloading. Service providers can dynamically update their caches to improve the system offloading efficiency. To maximize the offloading efficiency, we formulate the TSCU problem as a mixed-integer non-linear programming (MINLP) problem and develop an improved discrete cuckoo search algorithm to solve the optimization problem. Our proposed algorithm explores the global environment via Levy flight. On the other hand, the algorithm convergence rate is accelerated by generating the initial population based on a greedy algorithm and its variants and updating the discard probability. Simulation results show that our proposal has achieved an up to 12% improvement in offloading efficiency compared to the benchmark schemes.
- 25 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (9) , 12735-12747
UltraADV: An Unsupervised Deep Learning Lightweight Framework for Anomaly Detection in V2X
Designing anomaly detection systems for vehicle-to-everything (V2X) is a challenge. Deep learning has shown strong advantages in anomaly detection. However, labeling anomalies is often difficult and expensive, and deep neural networks often prioritize performances by incorporating a large number of parameters. Therefore, there is an urgent need for a lightweight anomaly detection framework which can be built with unsupervised learning methods. In this article, we propose a framework, the UltraADV, which is suitable for anomaly detection systems in V2X. In the framework, the autoencoder facilitates an unsupervised setting, while the knowledge distillation transfers the knowledge of a larger network to a smaller network. Ultimately, the compressed model can be deployed onto the resource-constrained devices for real-world applications in V2X. We evaluate the performances of multiple architectures of UltraADV on two publicly available datasets. Our best performing architecture demonstrably surpasses the benchmark in terms of performance metrics, parameter efficiency, and prediction speed.
- 25 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (9) , 12735-12747
UltraADV: An Unsupervised Deep Learning Lightweight Framework for Anomaly Detection in V2X
Designing anomaly detection systems for vehicle-to-everything (V2X) is a challenge. Deep learning has shown strong advantages in anomaly detection. However, labeling anomalies is often difficult and expensive, and deep neural networks often prioritize performances by incorporating a large number of parameters. Therefore, there is an urgent need for a lightweight anomaly detection framework which can be built with unsupervised learning methods. In this article, we propose a framework, the UltraADV, which is suitable for anomaly detection systems in V2X. In the framework, the autoencoder facilitates an unsupervised setting, while the knowledge distillation transfers the knowledge of a larger network to a smaller network. Ultimately, the compressed model can be deployed onto the resource-constrained devices for real-world applications in V2X. We evaluate the performances of multiple architectures of UltraADV on two publicly available datasets. Our best performing architecture demonstrably surpasses the benchmark in terms of performance metrics, parameter efficiency, and prediction speed.
- Open Access
- 8 October 2024
- Published by Springer Nature in Scientific Reports
- Vol. 14 (1) , 1-11
Dense projection fusion for 3D object detection
Fusing information from LiDAR and cameras can effectively enhance the overall perceptivity of autonomous vehicles in various scenarios. Despite the relatively good results achieved by point-wise fusion and Bird’s-Eye-View (BEV) fusion, they still cannot fully leverage the image information and lack of effective depth information. For any fusion methods, the multi-modal features first need to be concatenated along the channel, and then the fused features are extracted using convolutional layers. However, this type of fusion methods is effective, but too coarse which causes that the fused features cannot pay more attention to the regions with important features and suffer from severe noise. To tackle these issues, we propose in this paper a Dense Projection Fusion (DPFusion) approach. It consists of two new modules: dense depth map guided BEV transform (DGBT) module and multi-modal feature adaptive fusion (MFAF) module. The DGBT module first quickly estimates the depth of each pixel and then projects all image features to the BEV space, making full use of the image information. The MFAF module computes the image weights and point cloud weights for each channel in each BEV grid and then adaptively weights and fuses the image BEV features with the point cloud BEV features. It is worth pointing out that the MFAF module makes the fused features pay more attention to background outline and object outline. Our proposed DPFusion demonstrates competitive results in 3D object detection, achieving a mean Average Precision (mAP) of 70.4 and a nuScenes detection score (NDS) of 72.3 on the nuScenes validation set.
- 8 October 2024
- Published by Springer Nature in Scientific Reports
- Vol. 14 (1) , 1-11
Open AccessDense projection fusion for 3D object detection
Fusing information from LiDAR and cameras can effectively enhance the overall perceptivity of autonomous vehicles in various scenarios. Despite the relatively good results achieved by point-wise fusion and Bird’s-Eye-View (BEV) fusion, they still cannot fully leverage the image information and lack of effective depth information. For any fusion methods, the multi-modal features first need to be concatenated along the channel, and then the fused features are extracted using convolutional layers. However, this type of fusion methods is effective, but too coarse which causes that the fused features cannot pay more attention to the regions with important features and suffer from severe noise. To tackle these issues, we propose in this paper a Dense Projection Fusion (DPFusion) approach. It consists of two new modules: dense depth map guided BEV transform (DGBT) module and multi-modal feature adaptive fusion (MFAF) module. The DGBT module first quickly estimates the depth of each pixel and then projects all image features to the BEV space, making full use of the image information. The MFAF module computes the image weights and point cloud weights for each channel in each BEV grid and then adaptively weights and fuses the image BEV features with the point cloud BEV features. It is worth pointing out that the MFAF module makes the fused features pay more attention to background outline and object outline. Our proposed DPFusion demonstrates competitive results in 3D object detection, achieving a mean Average Precision (mAP) of 70.4 and a nuScenes detection score (NDS) of 72.3 on the nuScenes validation set.
- 23 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 25 (11) , 16831-16844
A Quantum-Resistant Key Management Scheme Using Blockchain in C-V2X
Cellular-Vehicle-to-Everything (C-V2X) is featured as a heterogenous network, with the support of cellular based uplink/downlink transmissions and sidelink communications as well as network backhauling, where infringement on security can happen. Additionally, the developing quantum attack strikes the original NP-hard problems to be solvable in polynomial time complexity. Therefore, it is fatal to build a secure environment for vehicles to communicate in C-V2X. In this paper, we propose a key management scheme utilizing blockchain deployed at the base station for distributed storage and retrieval of vehicles’ public keys. In the network, each vehicle is identified with a public-private key pair, upon which vehicular registration, authenticated key agreement and revocation rely. In particular, the lattice-based cryptography is applied to the distribution of the group key for broadcasting beacon messages and establishing session keys for a secure peer-to-peer (P2P) communication. Keys are updated as vehicles travel from one region to another and are revoked upon misbehavior detection. We analyzed the quantum resistance of our scheme theoretically and its conservation against certain attacks informally. Comparative analysis across computation, communication, storage and energy consumption demonstrates the characteristics of our scheme. Finally, the blockchain simulation demonstrates the scalability of our scheme.
- 23 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 25 (11) , 16831-16844
A Quantum-Resistant Key Management Scheme Using Blockchain in C-V2X
Cellular-Vehicle-to-Everything (C-V2X) is featured as a heterogenous network, with the support of cellular based uplink/downlink transmissions and sidelink communications as well as network backhauling, where infringement on security can happen. Additionally, the developing quantum attack strikes the original NP-hard problems to be solvable in polynomial time complexity. Therefore, it is fatal to build a secure environment for vehicles to communicate in C-V2X. In this paper, we propose a key management scheme utilizing blockchain deployed at the base station for distributed storage and retrieval of vehicles’ public keys. In the network, each vehicle is identified with a public-private key pair, upon which vehicular registration, authenticated key agreement and revocation rely. In particular, the lattice-based cryptography is applied to the distribution of the group key for broadcasting beacon messages and establishing session keys for a secure peer-to-peer (P2P) communication. Keys are updated as vehicles travel from one region to another and are revoked upon misbehavior detection. We analyzed the quantum resistance of our scheme theoretically and its conservation against certain attacks informally. Comparative analysis across computation, communication, storage and energy consumption demonstrates the characteristics of our scheme. Finally, the blockchain simulation demonstrates the scalability of our scheme.
- 22 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 11 (12) , 22232-22242
CTPP: A Conditional Trajectory Privacy Preservation Scheme Using Blockchain in V2X
In Vehicle-to-Everything (V2X), the periodic broadcasting and reception of safety beacons by vehicles are pivotal for ensuring security. Nevertheless, the transmission of these messages in unencrypted formats over public channels presents substantial security risks. Adversaries with prior knowledge can potentially reconstruct the trajectories of targeted vehicles, thereby compromising sensitive information such as the behavioral patterns of drivers. To combat this issue, this article proposes a novel blockchain-based conditional trajectory privacy preservation (CTPP) scheme, which consists of two primary components: first, an attack-and-defense model is designed where a message blurring technique is utilized to resist the correlational background knowledge attack; second, the blockchain is introduced to enable the decentralized storage of beacon messages, thereby facilitating CTPP. Additionally, we develop two novel metrics, namely, the utility loss and the privacy loss, to evaluate the effectiveness of our CTPP. The former is quantified as the disparity between data sets pre and post the application of the privacy preservation mechanism. The latter is calculated as the proportion of the longest common subtrajectory (LCST) length to the original trajectory length. Extensive experimental results demonstrate that our CTPP significantly enhances trajectory privacy while incurring a moderate utility loss compared to the existing methods.
- 22 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 11 (12) , 22232-22242
CTPP: A Conditional Trajectory Privacy Preservation Scheme Using Blockchain in V2X
In Vehicle-to-Everything (V2X), the periodic broadcasting and reception of safety beacons by vehicles are pivotal for ensuring security. Nevertheless, the transmission of these messages in unencrypted formats over public channels presents substantial security risks. Adversaries with prior knowledge can potentially reconstruct the trajectories of targeted vehicles, thereby compromising sensitive information such as the behavioral patterns of drivers. To combat this issue, this article proposes a novel blockchain-based conditional trajectory privacy preservation (CTPP) scheme, which consists of two primary components: first, an attack-and-defense model is designed where a message blurring technique is utilized to resist the correlational background knowledge attack; second, the blockchain is introduced to enable the decentralized storage of beacon messages, thereby facilitating CTPP. Additionally, we develop two novel metrics, namely, the utility loss and the privacy loss, to evaluate the effectiveness of our CTPP. The former is quantified as the disparity between data sets pre and post the application of the privacy preservation mechanism. The latter is calculated as the proportion of the longest common subtrajectory (LCST) length to the original trajectory length. Extensive experimental results demonstrate that our CTPP significantly enhances trajectory privacy while incurring a moderate utility loss compared to the existing methods.
- Open Access
- 5 January 2024
- Published by MDPI AG in Electronics
- Vol. 13 (2) , 258
A Survey of Physical Layer Secret Key Generation Enhanced by Intelligent Reflecting Surface
As wireless communication scenarios grow more complicated, security issues are becoming increasingly prominent and severe. In the Internet of Things and vehicle-to-everything scenarios, conventional cryptographic technology faces numerous challenges. These include difficulties in secret key distribution and management, low update rates of secret keys, and vulnerability to quantum attacks. Physical layer secret key generation is considered a promising solution to security issues. The perfect secrecy proposed by Shannon can be achieved by combining secret key generation and the one-time pad when the length of secret keys is equal to that of plaintext. Hence, it is important to increase secret key generation rates. Intelligent reflecting surfaces demonstrate great advantages in improving the secret key generation performance. This paper provides a comprehensive review of current research efforts related to secret key generation assisted by intelligent reflecting surfaces, which is divided into three main categories: introducing the randomness of intelligent reflecting surfaces, optimizing the reflecting coefficients, and designing probing protocols. Comparative results of existing optimization approaches are provided and discussed. Furthermore, we emphasize the significance of selecting a random source of secret key generation from the perspective of information theory. Finally, two significant application scenarios, the Industrial Internet of Things and vehicle-to-everything, are discussed, and some challenges and opportunities are presented.
- 5 January 2024
- Published by MDPI AG in Electronics
- Vol. 13 (2) , 258
Open AccessA Survey of Physical Layer Secret Key Generation Enhanced by Intelligent Reflecting Surface
As wireless communication scenarios grow more complicated, security issues are becoming increasingly prominent and severe. In the Internet of Things and vehicle-to-everything scenarios, conventional cryptographic technology faces numerous challenges. These include difficulties in secret key distribution and management, low update rates of secret keys, and vulnerability to quantum attacks. Physical layer secret key generation is considered a promising solution to security issues. The perfect secrecy proposed by Shannon can be achieved by combining secret key generation and the one-time pad when the length of secret keys is equal to that of plaintext. Hence, it is important to increase secret key generation rates. Intelligent reflecting surfaces demonstrate great advantages in improving the secret key generation performance. This paper provides a comprehensive review of current research efforts related to secret key generation assisted by intelligent reflecting surfaces, which is divided into three main categories: introducing the randomness of intelligent reflecting surfaces, optimizing the reflecting coefficients, and designing probing protocols. Comparative results of existing optimization approaches are provided and discussed. Furthermore, we emphasize the significance of selecting a random source of secret key generation from the perspective of information theory. Finally, two significant application scenarios, the Industrial Internet of Things and vehicle-to-everything, are discussed, and some challenges and opportunities are presented.
- 1 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in China Communications
- Vol. 21 (6) , 1-13
A probing model of secret key generation based on channel autocorrelation function
Secret key generation (SKG) is a promising solution to the problem of wireless communications security. As the first step of SKG, channel probing affects it significantly. Although there have been some probing schemes, there is a lack of research on the optimization of the probing process. This study investigates how to optimize correlated parameters to maximize the SKG rate (SKGR) in the time-division duplex (TDD) mode. First, we build a probing model which includes the effects of transmitting power, the probing period, and the dimension of sample vectors. Based on the model, the analytical expression of the SKGR is given. Next, we formulate an optimization problem for maximizing the SKGR and give an algorithm to solve it. We conclude the SKGR monotonically increases as the transmitting power increases. Relevant mathematical proofs are given in this study. From the simulation results, increasing appropriately the probing period and the dimension of the sample vector could increase the SKGR dramatically compared to a yardstick, which indicates the importance of optimizing the parameters related to the channel probing phase.
- 1 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in China Communications
- Vol. 21 (6) , 1-13
A probing model of secret key generation based on channel autocorrelation function
Secret key generation (SKG) is a promising solution to the problem of wireless communications security. As the first step of SKG, channel probing affects it significantly. Although there have been some probing schemes, there is a lack of research on the optimization of the probing process. This study investigates how to optimize correlated parameters to maximize the SKG rate (SKGR) in the time-division duplex (TDD) mode. First, we build a probing model which includes the effects of transmitting power, the probing period, and the dimension of sample vectors. Based on the model, the analytical expression of the SKGR is given. Next, we formulate an optimization problem for maximizing the SKGR and give an algorithm to solve it. We conclude the SKGR monotonically increases as the transmitting power increases. Relevant mathematical proofs are given in this study. From the simulation results, increasing appropriately the probing period and the dimension of the sample vector could increase the SKGR dramatically compared to a yardstick, which indicates the importance of optimizing the parameters related to the channel probing phase.
- 12 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Letters
- Vol. 28 (2) , 323-327
Energy and Gradient Aware Dynamic Scheduling for V2V Aided Federated Edge Learning
Federated edge learning (FEEL) is significant for effective machine learning (ML). However, the scarce communication resources and energy-limited vehicles result in part of the local gradient being available for model updates. A residual feedback mechanism is designed to fully use the local gradients. To enable more vehicles to upload the local gradients successfully under a given energy budget, vehicle-to-vehicle (V2V) is applied to explore the energy of neighbors and improve the communication condition to enhance the uploading. Then, the Lyapunov optimization method is employed. Simulation results show that the learning performance of the proposed algorithm outperforms other existing schemes.
- 12 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Letters
- Vol. 28 (2) , 323-327
Energy and Gradient Aware Dynamic Scheduling for V2V Aided Federated Edge Learning
Federated edge learning (FEEL) is significant for effective machine learning (ML). However, the scarce communication resources and energy-limited vehicles result in part of the local gradient being available for model updates. A residual feedback mechanism is designed to fully use the local gradients. To enable more vehicles to upload the local gradients successfully under a given energy budget, vehicle-to-vehicle (V2V) is applied to explore the energy of neighbors and improve the communication condition to enhance the uploading. Then, the Lyapunov optimization method is employed. Simulation results show that the learning performance of the proposed algorithm outperforms other existing schemes.