Yuhui Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yuhui Wang has a total of 12 co-authors. Here are the top co-authors by number of shared publications:
- Junaid Farooq
University of Michigan-Dearborn
5 shared publications - Gianluca Setti
Politecnico di Torino
3 shared publications - Hakim Ghazzai
Computer, Electrical and Mathematical Sciences and Engineering (CEMSE) Division, King Abdullah University of Science and Technology (KAUST), Thuwal, Saudi Arabia
3 shared publications Junaid Farooq
Department of Electrical and Computer Engineering, University of Michigan-Dearborn
2 shared publications- Juntao Chen
Department of Computer and Information Sciences, Fordham University, New York, USA
2 shared publications Xingqi Wu
Department of Electrical and Computer Engineering, University of Michigan-Dearborn
2 shared publications
Latest Publications
- 27 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Networking Letters
- Vol. 8, 199-203
Process-Informed Multi-Agent Reinforcement Learning for Joint Radio-Compute Orchestration in Industrial O-RAN
Industrial open radio access network (O-RAN) systems must support manufacturing workloads with dynamically varying communication and computation demands driven by production processes. Existing O-RAN orchestration mechanisms remain largely network-centric and fail to capture such process-induced variability. This letter proposes a decentralized process-informed reinforcement learning (RL) framework for joint radio and compute orchestration in industrial O-RAN. Distributed agents at radio units (RUs) adapt physical resource block (PRB) and compute allocations based on workload and latency requirements. Using a multi-agent dueling deep Q-network architecture, the approach improves latency fulfillment and reliability under nonstationary industrial traffic. Simulation results demonstrate consistent gains over network-centric multi-agent baselines.
- 27 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Networking Letters
- Vol. 8, 199-203
Process-Informed Multi-Agent Reinforcement Learning for Joint Radio-Compute Orchestration in Industrial O-RAN
Industrial open radio access network (O-RAN) systems must support manufacturing workloads with dynamically varying communication and computation demands driven by production processes. Existing O-RAN orchestration mechanisms remain largely network-centric and fail to capture such process-induced variability. This letter proposes a decentralized process-informed reinforcement learning (RL) framework for joint radio and compute orchestration in industrial O-RAN. Distributed agents at radio units (RUs) adapt physical resource block (PRB) and compute allocations based on workload and latency requirements. Using a multi-agent dueling deep Q-network architecture, the approach improves latency fulfillment and reliability under nonstationary industrial traffic. Simulation results demonstrate consistent gains over network-centric multi-agent baselines.
- Open Access
- 20 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Collaborative Cloud-Edge Computing via LLM-Guided Constrained Reinforcement Learning
Cloud computing (CC) and mobile edge computing (MEC) has become essential for augmenting computational capacity and minimizing latency by offloading tasks from user equipments (UEs) to either centralized cloud servers or proximal edge servers. While Deep Reinforcement Learning (DRL) is a powerful tool for optimizing offloading decisions and managing resource allocation, it often struggles as the network scales. The exponential expansion of the state-action space, combined with strict feasibility constraints, frequently leads to suboptimal policies and slow convergence in traditional DRL models. Large language models (LLMs) with strong reasoning capabilities and extensive prior knowledge offer a potential solution by enabling more efficient exploration and guiding the DRL toward better policies subject to system constraints. Therefore, we propose an LLM-guided constrained proximal policy optimization (LGC-PPO) algorithm that integrates LLM-generated policies, composed of action demonstrations and reward shaping. A probabilistic action sampling strategy is designed to control the action selection from LLM policy and PPO agent, which facilitate learning in high-dimensional decision spaces and improve earlystage decision quality. Moreover, the mixed strategy balances exploration and exploitation, contributing to stable and effective learning. Experimental results demonstrate that the proposed algorithm consistently outperforms baseline methods, particularly in large and complex decision spaces, highlighting its strong potential for resource optimization in collaborative cloud-edge computing networks.
- 20 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessCollaborative Cloud-Edge Computing via LLM-Guided Constrained Reinforcement Learning
Cloud computing (CC) and mobile edge computing (MEC) has become essential for augmenting computational capacity and minimizing latency by offloading tasks from user equipments (UEs) to either centralized cloud servers or proximal edge servers. While Deep Reinforcement Learning (DRL) is a powerful tool for optimizing offloading decisions and managing resource allocation, it often struggles as the network scales. The exponential expansion of the state-action space, combined with strict feasibility constraints, frequently leads to suboptimal policies and slow convergence in traditional DRL models. Large language models (LLMs) with strong reasoning capabilities and extensive prior knowledge offer a potential solution by enabling more efficient exploration and guiding the DRL toward better policies subject to system constraints. Therefore, we propose an LLM-guided constrained proximal policy optimization (LGC-PPO) algorithm that integrates LLM-generated policies, composed of action demonstrations and reward shaping. A probabilistic action sampling strategy is designed to control the action selection from LLM policy and PPO agent, which facilitate learning in high-dimensional decision spaces and improve earlystage decision quality. Moreover, the mixed strategy balances exploration and exploitation, contributing to stable and effective learning. Experimental results demonstrate that the proposed algorithm consistently outperforms baseline methods, particularly in large and complex decision spaces, highlighting its strong potential for resource optimization in collaborative cloud-edge computing networks.
- 17 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Networking Letters
- Vol. 7 (4) , 323-327
SliceWise: Traffic-Aware Multi-Agent RL with Temporal Encoding for Joint Slice Admission and Resource Orchestration in O-RAN
Ultra-reliable low-latency communication (URLLC) services in mission-critical applications such as industrial automation, remote healthcare, and autonomous systems demand stringent quality-of-service (QoS) guarantees under dynamic, resource-constrained wireless conditions. Open radio access networks (O-RAN) offer a software-defined and programmable foundation for embedding AI-driven control. This letter proposes SliceWise, a traffic-aware multi-agent deep reinforcement learning (MARL) framework for joint admission control, user-slice association, and resource orchestration in O-RAN. The control problem is modeled as a cooperative multi-agent Markov decision process, with long short-term memory (LSTM) networks capturing temporal traffic dynamics and dueling double deep Q-networks (D3QN) ensuring stable learning. Simulation results show that SliceWise substantially reduces delay violations and packet drops while improving slice-level reliability and resource efficiency, offering a scalable orchestration solution for AI-native O-RAN deployments.
- 17 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Networking Letters
- Vol. 7 (4) , 323-327
SliceWise: Traffic-Aware Multi-Agent RL with Temporal Encoding for Joint Slice Admission and Resource Orchestration in O-RAN
Ultra-reliable low-latency communication (URLLC) services in mission-critical applications such as industrial automation, remote healthcare, and autonomous systems demand stringent quality-of-service (QoS) guarantees under dynamic, resource-constrained wireless conditions. Open radio access networks (O-RAN) offer a software-defined and programmable foundation for embedding AI-driven control. This letter proposes SliceWise, a traffic-aware multi-agent deep reinforcement learning (MARL) framework for joint admission control, user-slice association, and resource orchestration in O-RAN. The control problem is modeled as a cooperative multi-agent Markov decision process, with long short-term memory (LSTM) networks capturing temporal traffic dynamics and dueling double deep Q-networks (D3QN) ensuring stable learning. Simulation results show that SliceWise substantially reduces delay violations and packet drops while improving slice-level reliability and resource efficiency, offering a scalable orchestration solution for AI-native O-RAN deployments.
- Open Access
- 14 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
ORCA: Offloading and Resource Coordination via Constrained Multi-Agent Reinforcement Learning in Dynamic Cloud-Edge Systems
The explosive growth of computation-intensive and latency-sensitive applications, such as augmented reality (AR), autonomous driving, and real-time analytics, imposes stringent computation, power, and bandwidth requirements on modern wireless networks. Traditional edge computing frameworks, despite offering reduced latency compared to cloud computing, often struggle to meet the massive and heterogeneous demands of such applications due to limited edge server resources and rapidly fluctuating network conditions. Collaborative and intelligent cloudedge computing systems have emerged as a promising paradigm to address these challenges by exploiting both the abundant computational resources of cloud servers and the low-latency advantages of edge servers. However, efficiently orchestrating task offloading and resource provisioning across the cloud and edge layers remains a significant challenge. Existing approaches based on static optimization or centralized deep learning often suffer from high computational complexity, lack of adaptability to real-time network dynamics, and scalability bottlenecks due to centralized information collection and decision-making. In this paper, we propose ORCA, a scalable framework for adaptive task offloading and resource provisioning in dynamic collaborative cloud-edge systems. ORCA leverages multi-agent proximal policy optimization (MAPPO) to enable decentralized, real-time decisionmaking by agents deployed at both cloud and edge servers. These agents learn to optimize task distribution, bandwidth allocation, and compute provisioning using only local observations, while coordinating to achieve global system efficiency. Extensive simulations demonstrate that ORCA dynamically adapts to varying task demands and server capabilities, and significantly outperforms baseline approaches in terms of energy efficiency, latency reduction, and system scalability.
- 14 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessORCA: Offloading and Resource Coordination via Constrained Multi-Agent Reinforcement Learning in Dynamic Cloud-Edge Systems
The explosive growth of computation-intensive and latency-sensitive applications, such as augmented reality (AR), autonomous driving, and real-time analytics, imposes stringent computation, power, and bandwidth requirements on modern wireless networks. Traditional edge computing frameworks, despite offering reduced latency compared to cloud computing, often struggle to meet the massive and heterogeneous demands of such applications due to limited edge server resources and rapidly fluctuating network conditions. Collaborative and intelligent cloudedge computing systems have emerged as a promising paradigm to address these challenges by exploiting both the abundant computational resources of cloud servers and the low-latency advantages of edge servers. However, efficiently orchestrating task offloading and resource provisioning across the cloud and edge layers remains a significant challenge. Existing approaches based on static optimization or centralized deep learning often suffer from high computational complexity, lack of adaptability to real-time network dynamics, and scalability bottlenecks due to centralized information collection and decision-making. In this paper, we propose ORCA, a scalable framework for adaptive task offloading and resource provisioning in dynamic collaborative cloud-edge systems. ORCA leverages multi-agent proximal policy optimization (MAPPO) to enable decentralized, real-time decisionmaking by agents deployed at both cloud and edge servers. These agents learn to optimize task distribution, bandwidth allocation, and compute provisioning using only local observations, while coordinating to achieve global system efficiency. Extensive simulations demonstrate that ORCA dynamically adapts to varying task demands and server capabilities, and significantly outperforms baseline approaches in terms of energy efficiency, latency reduction, and system scalability.
- 5 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (2) , 2320-2334
Dynamic Multi-Modal UAV Control for Optimized Coverage and Backhaul Connectivity in Spatially Unstructured and Dispersed User Environments
Unmanned aerial vehicles (UAVs) have emerged as a promising solution for establishing wireless communications in regions lacking terrestrial network infrastructure, such as remote or emergency areas. Deploying UAV networks effectively in these scenarios poses significant challenges due to the unknown and potentially complex locations of users. In scenarios where users are dispersed in intricate spatial patterns, achieving high coverage and resilient network connectivity among the UAV networks is challenging. The irregular and arbitrary distribution of users can lead to gaps in coverage, as traditional UAV placement optimization approaches are often unable to adapt to such dynamic environments. This complexity necessitates advanced strategies to ensure reliable and continuous network service to users. In this paper, we propose a distributed approach that leverages flocking dynamics and distributed consensus algorithms for dynamic UAV positioning. By enabling a multi-modal UAV operation policy, we develop a framework that enables the network to dynamically respond to complex user locations and establish backhaul connectivity between dispersed user clusters. Simulation results demonstrate that our approach successfully establishes a robust and adaptable UAV network capable of providing seamless coverage for complex user configurations and also ensuring comprehensive inter-cluster connectivity among dispersed user clusters. Additionally, the network exhibits strong resilience against random failures, swiftly recovering from disruptions to ensure stable and reliable communication even when UAVs are compromised.
- 5 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (2) , 2320-2334
Dynamic Multi-Modal UAV Control for Optimized Coverage and Backhaul Connectivity in Spatially Unstructured and Dispersed User Environments
Unmanned aerial vehicles (UAVs) have emerged as a promising solution for establishing wireless communications in regions lacking terrestrial network infrastructure, such as remote or emergency areas. Deploying UAV networks effectively in these scenarios poses significant challenges due to the unknown and potentially complex locations of users. In scenarios where users are dispersed in intricate spatial patterns, achieving high coverage and resilient network connectivity among the UAV networks is challenging. The irregular and arbitrary distribution of users can lead to gaps in coverage, as traditional UAV placement optimization approaches are often unable to adapt to such dynamic environments. This complexity necessitates advanced strategies to ensure reliable and continuous network service to users. In this paper, we propose a distributed approach that leverages flocking dynamics and distributed consensus algorithms for dynamic UAV positioning. By enabling a multi-modal UAV operation policy, we develop a framework that enables the network to dynamically respond to complex user locations and establish backhaul connectivity between dispersed user clusters. Simulation results demonstrate that our approach successfully establishes a robust and adaptable UAV network capable of providing seamless coverage for complex user configurations and also ensuring comprehensive inter-cluster connectivity among dispersed user clusters. Additionally, the network exhibits strong resilience against random failures, swiftly recovering from disruptions to ensure stable and reliable communication even when UAVs are compromised.
- Top Cited Papers
- 21 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. PP (15361233) , 1-15
Joint Positioning and Computation Offloading in Multi-UAV MEC for Low Latency Applications: a Proximal Policy Optimization Approach
Multi-access edge computing (MEC) has emerged as a proven solution for reducing communication latency and enhancing user experience in delay-sensitive applications by offloading computation-intensive tasks to edge servers. In future networks, unmanned aerial vehicles (UAVs), with their flexible deployment and reliable communication capabilities, have the potential to be deployed as aerial MEC servers in areas lacking cellular infrastructure. However, the joint optimization of UAV placement and task offloading poses significant challenges due to the interdependence between communication latency, computational demands, and the resource limitations of UAVs. In this paper, we propose a novel joint optimization framework utilizing proximal policy optimization (PPO) to simultaneously address UAV placement and computation offloading in UAVenabled MEC networks. The framework dynamically adapts to changing network conditions, minimizing end-to-end latency while balancing computational loads and energy consumption. Extensive simulations demonstrate that the proposed PPO-based approach achieves superior performance compared to conventional optimization methods, with significant improvements in system latency, resource utilization, and network resilience. This work contributes scalable, adaptive solutions for UAV-assisted MEC networks in dynamic environments, enabling robust support for mission-critical and latency-sensitive applications.
- 21 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. PP (15361233) , 1-15
Top Cited PapersJoint Positioning and Computation Offloading in Multi-UAV MEC for Low Latency Applications: a Proximal Policy Optimization Approach
Multi-access edge computing (MEC) has emerged as a proven solution for reducing communication latency and enhancing user experience in delay-sensitive applications by offloading computation-intensive tasks to edge servers. In future networks, unmanned aerial vehicles (UAVs), with their flexible deployment and reliable communication capabilities, have the potential to be deployed as aerial MEC servers in areas lacking cellular infrastructure. However, the joint optimization of UAV placement and task offloading poses significant challenges due to the interdependence between communication latency, computational demands, and the resource limitations of UAVs. In this paper, we propose a novel joint optimization framework utilizing proximal policy optimization (PPO) to simultaneously address UAV placement and computation offloading in UAVenabled MEC networks. The framework dynamically adapts to changing network conditions, minimizing end-to-end latency while balancing computational loads and energy consumption. Extensive simulations demonstrate that the proposed PPO-based approach achieves superior performance compared to conventional optimization methods, with significant improvements in system latency, resource utilization, and network resilience. This work contributes scalable, adaptive solutions for UAV-assisted MEC networks in dynamic environments, enabling robust support for mission-critical and latency-sensitive applications.
- 19 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 11 (8) , 14727-14738
Deep-Reinforcement-Learning-Based Placement for Integrated Access Backhauling in UAV-Assisted Wireless Networks
The advent of fifth generation 5G networks has opened new avenues for enhancing connectivity, particularly in challenging environments, such as remote areas or disaster-struck regions. Unmanned aerial vehicles (UAVs) have been identified as a versatile tool in this context, particularly for improving network performance through the integrated access and backhaul (IAB) feature of 5G. However, existing approaches to UAV-assisted network enhancement face limitations in dynamically adapting to varying user locations and network demands. This article introduces a novel approach leveraging deep reinforcement learning (DRL) to optimize UAV placement in real time, dynamically adjusting to changing network conditions and user requirements. Our method focuses on the intricate balance between fronthaul and backhaul links, a critical aspect often overlooked in current solutions. The unique contribution of this work lies in its ability to autonomously position UAVs in a way that not only ensures robust connectivity to ground users but also maintains seamless integration with central network infrastructure. Through various simulated scenarios, we demonstrate how our approach effectively addresses these challenges, enhancing coverage and network performance in critical areas. This research fills a significant gap in UAV-assisted 5G networks, providing a scalable and adaptive solution for future mobile networks.
- 19 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 11 (8) , 14727-14738
Deep-Reinforcement-Learning-Based Placement for Integrated Access Backhauling in UAV-Assisted Wireless Networks
The advent of fifth generation 5G networks has opened new avenues for enhancing connectivity, particularly in challenging environments, such as remote areas or disaster-struck regions. Unmanned aerial vehicles (UAVs) have been identified as a versatile tool in this context, particularly for improving network performance through the integrated access and backhaul (IAB) feature of 5G. However, existing approaches to UAV-assisted network enhancement face limitations in dynamically adapting to varying user locations and network demands. This article introduces a novel approach leveraging deep reinforcement learning (DRL) to optimize UAV placement in real time, dynamically adjusting to changing network conditions and user requirements. Our method focuses on the intricate balance between fronthaul and backhaul links, a critical aspect often overlooked in current solutions. The unique contribution of this work lies in its ability to autonomously position UAVs in a way that not only ensures robust connectivity to ground users but also maintains seamless integration with central network infrastructure. Through various simulated scenarios, we demonstrate how our approach effectively addresses these challenges, enhancing coverage and network performance in critical areas. This research fills a significant gap in UAV-assisted 5G networks, providing a scalable and adaptive solution for future mobile networks.