Yufeng Zhan
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Yufeng Zhan has a total of 92 co-authors. Here are the top co-authors by number of shared publications:
- Yuanqing Xia
Zhongyuan University of Technology
31 shared publications - Di-Hua Zhai
Beijing Institute of Technology
7 shared publications - Song Guo
Hong Kong University of Science and Technology, Hong Kong, China
7 shared publications - Zicong Hong
Hong Kong University of Science and Technology, Hong Kong, Hong Kong
5 shared publications Haosong Peng
School of Automation Beijing Institute of Technology Beijing China
4 shared publications- Yi Liu
Department of Computing, The Hong Kong Polytechnic University, Hong Kong, SAR, China
4 shared publications
Latest Publications
- 2 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computers
- Vol. 75  (8) , 2915-2928
Patch-of-Interest ViT Inference Acceleration System for Edge-Assisted Video Analytics
The advent of edge computing has made real-time intelligent video analytics feasible. Previous works, based on traditional model architecture (e.g., CNN, RNN, etc.), employ various strategies to filter out non-region-of-interest content to minimize bandwidth and computation consumption but show inferior performance in adverse environments. Recently, visual foundation models based on transformers have shown great performance in adverse environments due to their amazing generalization capability. However, they require a large amount of computation power, which limits their applications in realtime intelligent video analytics. In this paper, we find visual foundation models like Vision Transformer (ViT) also have a dedicated acceleration mechanism for video analytics. To this end, we introduce Arena, an end-to-end edge-assisted video inference acceleration system based on ViT. We leverage the capability of ViT that can be accelerated through token pruning by only offloading and feeding Patches-of-Interest to the downstream models. Additionally, we design an adaptive keyframe inference switching algorithm tailored to different videos, capable of adapting to the current video content to jointly optimize accuracy and bandwidth. Through extensive experiments, our findings reveal that Arena can boost inference speeds by up to 1.58×, 1.82× and 1.98× on average while consuming only 47%, 31% and 27% of the bandwidth, respectively, all with high inference accuracy.
- 2 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computers
- Vol. 75  (8) , 2915-2928
Patch-of-Interest ViT Inference Acceleration System for Edge-Assisted Video Analytics
The advent of edge computing has made real-time intelligent video analytics feasible. Previous works, based on traditional model architecture (e.g., CNN, RNN, etc.), employ various strategies to filter out non-region-of-interest content to minimize bandwidth and computation consumption but show inferior performance in adverse environments. Recently, visual foundation models based on transformers have shown great performance in adverse environments due to their amazing generalization capability. However, they require a large amount of computation power, which limits their applications in realtime intelligent video analytics. In this paper, we find visual foundation models like Vision Transformer (ViT) also have a dedicated acceleration mechanism for video analytics. To this end, we introduce Arena, an end-to-end edge-assisted video inference acceleration system based on ViT. We leverage the capability of ViT that can be accelerated through token pruning by only offloading and feeding Patches-of-Interest to the downstream models. Additionally, we design an adaptive keyframe inference switching algorithm tailored to different videos, capable of adapting to the current video content to jointly optimize accuracy and bandwidth. Through extensive experiments, our findings reveal that Arena can boost inference speeds by up to 1.58×, 1.82× and 1.98× on average while consuming only 47%, 31% and 27% of the bandwidth, respectively, all with high inference accuracy.
- 9 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Services Computing
- Vol. 19  (2) , 1506-1519
Radiant: Efficient Timely Large-Scale Scene Analytics Based on Hierarchical Framework
With the advancement of computer vision, the recently emerged 3D Gaussian Splatting (3DGS) has increasingly become a popular scene analytics algorithm due to its outstanding performance. Existing cloud-based 3DGS architectures overlook the challenges in real-world environments when handling large-scale scene analysis. This exposes issues such as inefficiency, low security, lack of privacy, and limited scalability. In this paper, we propose Radiant, a hierarchical framework for large scene analytics in a heterogeneous cloud-edge-device system, which jointly considers high efficiency, privacy and security, and scalability. Via extensive empirical study, we find that it is crucial to partition the regions for each edge appropriately and allocate varying camera positions to each device for image collection and training. The core of Radiant is partitioning regions based on heterogeneous environment information and allocating workloads to each device accordingly. Furthermore, we provide a 3DGS model aggregation algorithm that enhances the quality and ensures the continuity of models' boundaries. Finally, we develop a testbed, and experiments demonstrate that Radiant improved reconstruction quality by up to 25.7% and reduced up to 79.6% end-to-end latency.
- 9 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Services Computing
- Vol. 19  (2) , 1506-1519
Radiant: Efficient Timely Large-Scale Scene Analytics Based on Hierarchical Framework
With the advancement of computer vision, the recently emerged 3D Gaussian Splatting (3DGS) has increasingly become a popular scene analytics algorithm due to its outstanding performance. Existing cloud-based 3DGS architectures overlook the challenges in real-world environments when handling large-scale scene analysis. This exposes issues such as inefficiency, low security, lack of privacy, and limited scalability. In this paper, we propose Radiant, a hierarchical framework for large scene analytics in a heterogeneous cloud-edge-device system, which jointly considers high efficiency, privacy and security, and scalability. Via extensive empirical study, we find that it is crucial to partition the regions for each edge appropriately and allocate varying camera positions to each device for image collection and training. The core of Radiant is partitioning regions based on heterogeneous environment information and allocating workloads to each device accordingly. Furthermore, we provide a 3DGS model aggregation algorithm that enhances the quality and ensures the continuity of models' boundaries. Finally, we develop a testbed, and experiments demonstrate that Radiant improved reconstruction quality by up to 25.7% and reduced up to 79.6% end-to-end latency.
- 18 February 2026
- Published by Elsevier in Future Generation Computer Systems
- Vol. 181
MCFL: A federated learning framework for network traffic classification in dynamic networks
- 18 February 2026
- Published by Elsevier in Future Generation Computer Systems
- Vol. 181
MCFL: A federated learning framework for network traffic classification in dynamic networks
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Automation Science and Engineering
- Vol. 23, 15238-15255
Disturbance-Rejection Reinforcement Learning for Control of Unknown Nonlinear Systems
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Automation Science and Engineering
- Vol. 23, 15238-15255
Disturbance-Rejection Reinforcement Learning for Control of Unknown Nonlinear Systems
- 1 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Journal of Systems Engineering and Electronics
- Vol. 36  (6) , 1665-1678
Differential Flatness ADRC for High-Speed Steering of Tracked Tank Systems
This paper proposes a differential-fatness-based active disturbance rejection control (ADRC) for high-speed steering control of tracked tank systems. Firstly, a high-speed steering model is established by considering the lateral component of the centrifugal force acting on the tank on the basis of modeling and analyzing the dynamic model of the low-speed steering system. Secondly, we propose a differential-flatness ADRC approach by converting the under-actuated system to a fully driven flat one. Moreover, we prove the differential flatness of the steering system, which facilitates a two-channel ADRC development. Finally, we show that both the states of the flat system and the original under-actuated system can track the reference trajectory. On the external interference condition, the system is observed to re-track the target signal within 2 s.
- 1 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Journal of Systems Engineering and Electronics
- Vol. 36  (6) , 1665-1678
Differential Flatness ADRC for High-Speed Steering of Tracked Tank Systems
This paper proposes a differential-fatness-based active disturbance rejection control (ADRC) for high-speed steering control of tracked tank systems. Firstly, a high-speed steering model is established by considering the lateral component of the centrifugal force acting on the tank on the basis of modeling and analyzing the dynamic model of the low-speed steering system. Secondly, we propose a differential-flatness ADRC approach by converting the under-actuated system to a fully driven flat one. Moreover, we prove the differential flatness of the steering system, which facilitates a two-channel ADRC development. Finally, we show that both the states of the flat system and the original under-actuated system can track the reference trajectory. On the external interference condition, the system is observed to re-track the target signal within 2 s.
- Open Access
- 19 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 1679-1693
Sylva: Tailoring Personalized Adversarial Defense in Pre-trained Models via Collaborative Fine-tuning
- 19 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 1679-1693
Open AccessSylva: Tailoring Personalized Adversarial Defense in Pre-trained Models via Collaborative Fine-tuning
- 27 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Dependable and Secure Computing
- Vol. 23  (2) , 2257-2274
Scaling Blockchain via Dynamic Sharding
Sharding is considered a promising solution for scaling blockchain systems. However, most existing sharding systems have not considered the dynamics of the environment when making a sharding strategy, including the change of pending transactions, the leaving and joining of participants, and malicious attacks, which could cause performance instability and security issues. To address it, in this paper, we propose an intelligent and efficient dynamic sharding technology to advance the blockchain system performance and security. We first propose a formal and general evaluation framework for blockchain sharding in a dynamic environment, and conclude an optimization target for the system performance and security. To achieve a long-term benefit for the optimization target, a deep reinforcement learning (DRL)-based sharding approach has been proposed to intelligently make optimal sharding strategies. Next, we propose an adaptive resharding protocol to efficiently reduce the overhead introduced by dynamic sharding. Our experimental results illustrate that our proposed dynamic sharding in a simulation testbed can achieve 2.8 times transactions per second compared to traditional static sharding systems, and guarantee high security in a dynamic environment.
- 27 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Dependable and Secure Computing
- Vol. 23  (2) , 2257-2274
Scaling Blockchain via Dynamic Sharding
Sharding is considered a promising solution for scaling blockchain systems. However, most existing sharding systems have not considered the dynamics of the environment when making a sharding strategy, including the change of pending transactions, the leaving and joining of participants, and malicious attacks, which could cause performance instability and security issues. To address it, in this paper, we propose an intelligent and efficient dynamic sharding technology to advance the blockchain system performance and security. We first propose a formal and general evaluation framework for blockchain sharding in a dynamic environment, and conclude an optimization target for the system performance and security. To achieve a long-term benefit for the optimization target, a deep reinforcement learning (DRL)-based sharding approach has been proposed to intelligently make optimal sharding strategies. Next, we propose an adaptive resharding protocol to efficiently reduce the overhead introduced by dynamic sharding. Our experimental results illustrate that our proposed dynamic sharding in a simulation testbed can achieve 2.8 times transactions per second compared to traditional static sharding systems, and guarantee high security in a dynamic environment.
- 10 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Electronics
- Vol. 73  (2) , 3025-3036
Unsupervised Cross-Domain Attack Traffic Classifier for Intelligent Connected Vehicle
The heterogeneous communication ecosystem in intelligent connected vehicles (ICVs) has significantly broadened attack surfaces, making accurate attack traffic classification crucial for the security of vehicular networks. Existing deep learning (DL)-based classification methods face challenges such as reliance on labeled data and susceptibility to cross-domain data distribution discrepancies. To mitigate these challenges, a cross-domain attack traffic classifier named domain adaptation noisy label (DANL) is proposed in this article. DANL integrates uncertainty-aware sample selection and graph spectral alignment into unsupervised domain adaptation (UDA) frameworks to bridge domain gaps. Furthermore, a dual correction framework integrating noisy label learning (LNL) with pseudolabel aggregation is developed to further correct misclassifications caused by erroneous alignment. Extensive experiments on public datasets demonstrate DANL’s superiority, achieving an improvement of classification accuracy up to 5.8% over baselines. Further validation in real-world tests on self-constructed ICV testbed with simulated attacks confirming its effectiveness.
- 10 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Electronics
- Vol. 73  (2) , 3025-3036
Unsupervised Cross-Domain Attack Traffic Classifier for Intelligent Connected Vehicle
The heterogeneous communication ecosystem in intelligent connected vehicles (ICVs) has significantly broadened attack surfaces, making accurate attack traffic classification crucial for the security of vehicular networks. Existing deep learning (DL)-based classification methods face challenges such as reliance on labeled data and susceptibility to cross-domain data distribution discrepancies. To mitigate these challenges, a cross-domain attack traffic classifier named domain adaptation noisy label (DANL) is proposed in this article. DANL integrates uncertainty-aware sample selection and graph spectral alignment into unsupervised domain adaptation (UDA) frameworks to bridge domain gaps. Furthermore, a dual correction framework integrating noisy label learning (LNL) with pseudolabel aggregation is developed to further correct misclassifications caused by erroneous alignment. Extensive experiments on public datasets demonstrate DANL’s superiority, achieving an improvement of classification accuracy up to 5.8% over baselines. Further validation in real-world tests on self-constructed ICV testbed with simulated attacks confirming its effectiveness.