Qinghai Liu
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
Qinghai Liu has a total of 40 co-authors. Here are the top co-authors by number of shared publications:
- Lun Tang
School of Communications and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing, China
7 shared publications - Qianbin Chen
Chongqing University of Posts and Telecommunications
6 shared publications Cho-Lea Tso
2 shared publications- Jason Liu
City of Hope National Medical Center
2 shared publications - Jianyong Yang
School of Intelligent Manufacturing, Chongqing Vocational College of Intelligent Engineering, Chongqing, China
2 shared publications - Koji Yoshimoto
Department of Neurosurgery, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan
2 shared publications
Latest Publications
- 21 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (8) , 15466-15484
LLM-PD: A Large Language Model-Driven Policy Distillation-Based Method for Multi-UAV Path Planning
The Industrial Internet of Things (IIoT) has accelerated the adoption of multi-uncrewed aerial vehicle (UAV) systems in applications, such as urban inspection and emergency response. However, the effective path planning in dynamic environments remains challenging due to slow convergence, weak coordination, trajectory oscillation, and limited generalization. To address these issues, this article proposes a large language model-driven policy distillation (LLM-PD)-based method for multi-UAV path planning. By integrating LLM-based semantic reasoning with multiagent reinforcement learning (MARL), LLM-PD introduces a three-stage training paradigm—policy distillation, imitation pretraining, and reinforcement-based self-optimization—to enhance the policy quality and generalization. The problem is formulated as a partially observable Markov decision process (POMDP). A teacher model combining a dynamic graph attention network (GAT) and a large language model (LLM) is designed, where the GAT constructs a dynamic heterogeneous topology of UAVs, obstacles, and targets, while the LLM performs semantic reasoning to generate coordinated expert strategies. A joint action evaluation mechanism further builds a high-value expert knowledge base. The student model adopts a hybrid imitation–reinforcement framework. Imitation learning (IL) first clones expert behaviors from the knowledge base to accelerate policy acquisition, after which a centralized training with decentralized execution (CTDE)-based PPO stage refines the policy using interaction data, improving adaptability to wind disturbances and dynamic obstacles. This unified framework enables an effective transition from expert-guided learning to autonomous optimization. Simulation results show that LLM-PD substantially outperforms state-of-the-art baselines in path efficiency, convergence speed, cooperative stability, and generalization. Ablation studies further confirm the complementary benefits of LLM reasoning, GAT structural perception, and the imitation–reinforcement integration mechanism.
- 21 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (8) , 15466-15484
LLM-PD: A Large Language Model-Driven Policy Distillation-Based Method for Multi-UAV Path Planning
The Industrial Internet of Things (IIoT) has accelerated the adoption of multi-uncrewed aerial vehicle (UAV) systems in applications, such as urban inspection and emergency response. However, the effective path planning in dynamic environments remains challenging due to slow convergence, weak coordination, trajectory oscillation, and limited generalization. To address these issues, this article proposes a large language model-driven policy distillation (LLM-PD)-based method for multi-UAV path planning. By integrating LLM-based semantic reasoning with multiagent reinforcement learning (MARL), LLM-PD introduces a three-stage training paradigm—policy distillation, imitation pretraining, and reinforcement-based self-optimization—to enhance the policy quality and generalization. The problem is formulated as a partially observable Markov decision process (POMDP). A teacher model combining a dynamic graph attention network (GAT) and a large language model (LLM) is designed, where the GAT constructs a dynamic heterogeneous topology of UAVs, obstacles, and targets, while the LLM performs semantic reasoning to generate coordinated expert strategies. A joint action evaluation mechanism further builds a high-value expert knowledge base. The student model adopts a hybrid imitation–reinforcement framework. Imitation learning (IL) first clones expert behaviors from the knowledge base to accelerate policy acquisition, after which a centralized training with decentralized execution (CTDE)-based PPO stage refines the policy using interaction data, improving adaptability to wind disturbances and dynamic obstacles. This unified framework enables an effective transition from expert-guided learning to autonomous optimization. Simulation results show that LLM-PD substantially outperforms state-of-the-art baselines in path efficiency, convergence speed, cooperative stability, and generalization. Ablation studies further confirm the complementary benefits of LLM reasoning, GAT structural perception, and the imitation–reinforcement integration mechanism.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (6) , 12437-12460
Digital Twin-Assisted VNF Migration Algorithm Based on Spatiotemporal Large Model Resource Demand Prediction
To address the issues of lagging virtual network function (VNF) migration caused by the dynamic changes in resource requirements in Industrial Internet of Things (IIoT) and the unsatisfactory prediction effect due to the neglect of the connection relationships between nodes in resource requirement prediction, a digital twin(DT)-assisted VNF migration algorithm with spatiotemporal large language model-based resource demand prediction is proposed. Firstly, a large language model-based resource prediction method that combines graph convolutional networks and multi-head self-attention mechanism is introduced to effectively capture spatio-temporal correlations and predict future resource requirements. Next, in order to ensure a deterministic quality of service (QoS) and optimize migration decisions, a DT-assisted VNF migration model composed of energy consumption, latency, and load balancing is constructed, and a joint optimization model for VNF migration and DT re-association aimed at maximizing the long-term utility of the system is established. Finally, a migration algorithm based on heterogeneous multi-agent proximal policy optimization is proposed to solve the VNF migration problem according to the resource requirements predicted by the model. To address the problem of excessive synchronization delays of DT nodes originally associated with nodes after migration, a counterfactual multi-agent algorithm is proposed to solve the problem of DT re-association. Simulation results show that the proposed algorithm improves prediction accuracy, ensures load balancing, and reduces system energy consumption and synchronization delays.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (6) , 12437-12460
Digital Twin-Assisted VNF Migration Algorithm Based on Spatiotemporal Large Model Resource Demand Prediction
To address the issues of lagging virtual network function (VNF) migration caused by the dynamic changes in resource requirements in Industrial Internet of Things (IIoT) and the unsatisfactory prediction effect due to the neglect of the connection relationships between nodes in resource requirement prediction, a digital twin(DT)-assisted VNF migration algorithm with spatiotemporal large language model-based resource demand prediction is proposed. Firstly, a large language model-based resource prediction method that combines graph convolutional networks and multi-head self-attention mechanism is introduced to effectively capture spatio-temporal correlations and predict future resource requirements. Next, in order to ensure a deterministic quality of service (QoS) and optimize migration decisions, a DT-assisted VNF migration model composed of energy consumption, latency, and load balancing is constructed, and a joint optimization model for VNF migration and DT re-association aimed at maximizing the long-term utility of the system is established. Finally, a migration algorithm based on heterogeneous multi-agent proximal policy optimization is proposed to solve the VNF migration problem according to the resource requirements predicted by the model. To address the problem of excessive synchronization delays of DT nodes originally associated with nodes after migration, a counterfactual multi-agent algorithm is proposed to solve the problem of DT re-association. Simulation results show that the proposed algorithm improves prediction accuracy, ensures load balancing, and reduces system energy consumption and synchronization delays.
- 22 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (24) , 52748-52765
Digital Twin-Based Joint Optimization Strategy for Dual Time-Domain Slice Resource Management and DT Deployment in IoV
To address the diverse user service requirements in network slicing (NS) and the challenges of synchronization accuracy and low latency in digital twin (DT) deployment, this article proposes a joint optimization strategy for DT-based dual time-domain slice resource management and DT deployment in the Internet of Vehicles (IoV). First, a demand-driven slice resource management scheme is introduced to mitigate quality of service (QoS) degradation caused by insufficient resource allocation under fluctuating user demands. Network performance indicator weights are dynamically adjusted using demand enhancement factors. Second, to minimize the impact of DT synchronization on resource allocation and address challenges like low latency and delay bias, DT utility is quantified from three perspectives: completeness, load contribution, and resource reliability. Finally, a price incentive mechanism with dynamic load adjustment is designed to balance the supply and demand of DT and service resources. This joint optimization problem is an NP-hard mixed-integer nonlinear problem with dual time-domain coupling, which can be decomposed into utility maximization strategies in different time domains. In the long-time domain (LTD), a Q-value-based deep transfer reinforcement learning (QDTRL) algorithm is used for DT deployment, while in the short-time domain (STD), a long short-term memory multi-agent proximal policy optimization (LSTM-MAPPO) algorithm is applied for resource allocation and DT synchronization weight adjustment. Simulation results show that, compared to baseline schemes, the proposed strategy achieves higher utility, accelerates convergence, and effectively allocates resources and synchronizes DT.
- 22 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (24) , 52748-52765
Digital Twin-Based Joint Optimization Strategy for Dual Time-Domain Slice Resource Management and DT Deployment in IoV
To address the diverse user service requirements in network slicing (NS) and the challenges of synchronization accuracy and low latency in digital twin (DT) deployment, this article proposes a joint optimization strategy for DT-based dual time-domain slice resource management and DT deployment in the Internet of Vehicles (IoV). First, a demand-driven slice resource management scheme is introduced to mitigate quality of service (QoS) degradation caused by insufficient resource allocation under fluctuating user demands. Network performance indicator weights are dynamically adjusted using demand enhancement factors. Second, to minimize the impact of DT synchronization on resource allocation and address challenges like low latency and delay bias, DT utility is quantified from three perspectives: completeness, load contribution, and resource reliability. Finally, a price incentive mechanism with dynamic load adjustment is designed to balance the supply and demand of DT and service resources. This joint optimization problem is an NP-hard mixed-integer nonlinear problem with dual time-domain coupling, which can be decomposed into utility maximization strategies in different time domains. In the long-time domain (LTD), a Q-value-based deep transfer reinforcement learning (QDTRL) algorithm is used for DT deployment, while in the short-time domain (STD), a long short-term memory multi-agent proximal policy optimization (LSTM-MAPPO) algorithm is applied for resource allocation and DT synchronization weight adjustment. Simulation results show that, compared to baseline schemes, the proposed strategy achieves higher utility, accelerates convergence, and effectively allocates resources and synchronizes DT.
- 4 June 2025
- Published by Elsevier in Engineering Applications of Artificial Intelligence
- Vol. 157
Multi-modal semantic feature alignment medical cross-modal hashing
- 4 June 2025
- Published by Elsevier in Engineering Applications of Artificial Intelligence
- Vol. 157
Multi-modal semantic feature alignment medical cross-modal hashing
- Open Access
- 14 June 2024
- Published by Institution of Engineering and Technology (IET) in IET Image Processing
- Vol. 18 (10) , 2702-2714
TanrsColour: Transformer‐based medical image colourization with content and structure preservation
Medical image colouring techniques enable to colourize grey‐scale medical images for assisting doctors in diagnosis. Benefiting from the non‐linear fitting ability of deep neural network, deep medical image colouring techniques have achieved remarkable results. However, existing methods are still facing content and structure feature leakage, unrealistic colouring and poor scale invariability. Thus, this paper, proposes a Transformer‐based medical image colouring algorithm with long‐term dependency to avoid feature leakage of coloured images. To be specific, this method employs two different Transformer encoders to generate and encode feature sequences for grey‐scale medical images and real human colour slice images, respectively. Then, a novel multi‐layer Transformer decoder is used to stylize grey‐scale map image features based on the real physical colour feature sequences. For colouring images at different scales, we implement content‐ aware positional encoding with scale invariance and propose style‐aware positional encoding strategy to take realistic and physical colour prior into account. Extensive experimental results indicate our method has achieved better colourization effects than recent state‐of‐the‐art medical image colourization methods.
- 14 June 2024
- Published by Institution of Engineering and Technology (IET) in IET Image Processing
- Vol. 18 (10) , 2702-2714
Open AccessTanrsColour: Transformer‐based medical image colourization with content and structure preservation
Medical image colouring techniques enable to colourize grey‐scale medical images for assisting doctors in diagnosis. Benefiting from the non‐linear fitting ability of deep neural network, deep medical image colouring techniques have achieved remarkable results. However, existing methods are still facing content and structure feature leakage, unrealistic colouring and poor scale invariability. Thus, this paper, proposes a Transformer‐based medical image colouring algorithm with long‐term dependency to avoid feature leakage of coloured images. To be specific, this method employs two different Transformer encoders to generate and encode feature sequences for grey‐scale medical images and real human colour slice images, respectively. Then, a novel multi‐layer Transformer decoder is used to stylize grey‐scale map image features based on the real physical colour feature sequences. For colouring images at different scales, we implement content‐ aware positional encoding with scale invariance and propose style‐aware positional encoding strategy to take realistic and physical colour prior into account. Extensive experimental results indicate our method has achieved better colourization effects than recent state‐of‐the‐art medical image colourization methods.
- 5 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 10 (22) , 19825-19838
Federated Continuous Learning Based on Stacked Broad Learning System Assisted by Digital Twin Networks: An Incremental Learning Approach for Intrusion Detection in UAV Networks
The edge of the Internet of Things (IoT), which consists of unmanned aerial vehicles (UAVs), is vulnerable to network intrusion because software and wireless connections are used extensively in the IoT. Designing an efficient intrusion detection system (IDS) model is imperative. However, when creating IDS models with distributed data collected by UAVs, it is necessary to take precautions to protect the data’s security and privacy. Furthermore, most of the IDS models are focused on one-time learning but not on continuous learning. To this end, we propose a federated continuous learning framework with a stacked broad learning system (FCL-SBLS) based on the digital twin network (DTN), which can learn and train the IDS model on new data quickly and continuously. In order to improve the efficiency and quality of the IDS model when training and aggregation, we employ an asynchronous federated learning (FL) architecture, and a deep deterministic policy gradient (DDPG)-based UAV selection scheme assisted by DTN is proposed to help the global IDS model aggregation. The presented algorithm is validated using the CIC-IDS2017 data set, and the simulation results reveal that our algorithm achieves higher efficiency and accuracy than the existing FL scheme.
- 5 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 10 (22) , 19825-19838
Federated Continuous Learning Based on Stacked Broad Learning System Assisted by Digital Twin Networks: An Incremental Learning Approach for Intrusion Detection in UAV Networks
The edge of the Internet of Things (IoT), which consists of unmanned aerial vehicles (UAVs), is vulnerable to network intrusion because software and wireless connections are used extensively in the IoT. Designing an efficient intrusion detection system (IDS) model is imperative. However, when creating IDS models with distributed data collected by UAVs, it is necessary to take precautions to protect the data’s security and privacy. Furthermore, most of the IDS models are focused on one-time learning but not on continuous learning. To this end, we propose a federated continuous learning framework with a stacked broad learning system (FCL-SBLS) based on the digital twin network (DTN), which can learn and train the IDS model on new data quickly and continuously. In order to improve the efficiency and quality of the IDS model when training and aggregation, we employ an asynchronous federated learning (FL) architecture, and a deep deterministic policy gradient (DDPG)-based UAV selection scheme assisted by DTN is proposed to help the global IDS model aggregation. The presented algorithm is validated using the CIC-IDS2017 data set, and the simulation results reveal that our algorithm achieves higher efficiency and accuracy than the existing FL scheme.
- 1 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 10 (21) , 19102-19116
Deep Reinforcement Learning for Resource Demand Prediction and Virtual Function Network Migration in Digital Twin Network
Internet of Things (IoT) enables intelligent services varying with the complex and realtime environment to achieve network benefits, where Network Function Virtualization (NFV) can dynamically provide Virtualized Network Functions (VNFs) for IoT devices. In NFV-enabled IoT architecture, a service function chain (SFC) consists of an ordered set of virtual network functions (VNFs). However, the energy consumption of the VNF migration and SFC reconfiguration is one major issue owing to the dynamic characteristic of IoT network. In this paper, we propose a new paradigm digital twin (DT) to create the virtual twin of physical objects in IoT network, then, we formalize the problem as a mathematical model, which aims to minimize the energy consumption. To this end, we prove this problem is NP-hard and propose an algorithm Bidirectional Gated Recurrent Unit (Bi-GRU) based on federated learning to predict the resource requirement. Further more, according to the prediction result, which utilizing the deep reinforcement learning (DRL) algorithm for decision making of the VNF migration. Simulation results show that our proposed method can effectively reduce the number of VNFs to be migrated and economize the energy consumption of the digital twin IoT network.
- 1 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 10 (21) , 19102-19116
Deep Reinforcement Learning for Resource Demand Prediction and Virtual Function Network Migration in Digital Twin Network
Internet of Things (IoT) enables intelligent services varying with the complex and realtime environment to achieve network benefits, where Network Function Virtualization (NFV) can dynamically provide Virtualized Network Functions (VNFs) for IoT devices. In NFV-enabled IoT architecture, a service function chain (SFC) consists of an ordered set of virtual network functions (VNFs). However, the energy consumption of the VNF migration and SFC reconfiguration is one major issue owing to the dynamic characteristic of IoT network. In this paper, we propose a new paradigm digital twin (DT) to create the virtual twin of physical objects in IoT network, then, we formalize the problem as a mathematical model, which aims to minimize the energy consumption. To this end, we prove this problem is NP-hard and propose an algorithm Bidirectional Gated Recurrent Unit (Bi-GRU) based on federated learning to predict the resource requirement. Further more, according to the prediction result, which utilizing the deep reinforcement learning (DRL) algorithm for decision making of the VNF migration. Simulation results show that our proposed method can effectively reduce the number of VNFs to be migrated and economize the energy consumption of the digital twin IoT network.
- Open Access
- 3 April 2023
- Published by American Association for Cancer Research (AACR)
Supplementary Table 2 from Primary Glioblastomas Express Mesenchymal Stem-Like Properties
- 3 April 2023
- Published by American Association for Cancer Research (AACR)
Open AccessSupplementary Table 2 from Primary Glioblastomas Express Mesenchymal Stem-Like Properties
- Open Access
- 3 April 2023
- Published by American Association for Cancer Research (AACR)
Supplementary Table 1 from Primary Glioblastomas Express Mesenchymal Stem-Like Properties
Supplementary Table 1 from Primary Glioblastomas Express Mesenchymal Stem-Like Properties
- 3 April 2023
- Published by American Association for Cancer Research (AACR)
Open AccessSupplementary Table 1 from Primary Glioblastomas Express Mesenchymal Stem-Like Properties
Supplementary Table 1 from Primary Glioblastomas Express Mesenchymal Stem-Like Properties
- 29 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 10 (15) , 13370-13384
Handoff Control and Resource Allocation for RAN Slicing in IoT Based on DTN: An Improved Algorithm Based on Actor–Critic Framework
As a three-layer association of Internet of things equipment (IoTE) -network slicing (NS) -base station (BS) in radio access network (RAN) slicing, handoff control and resource allocation has become an important but complicated issue. In addition, the centralized controller has a difficult grasping the network situation in real time. In view of this, the problem of handoff control in the RAN slicing are investigated in the digital twin network (DTN), with the goal of maximizing the long-term utility about user satisfaction and handoff cost. Then an improved algorithm based on actor-critic framework is suggested, which is called HCRA. Specifically, the actor component contains neural networks for Handoff Control and an optimizer for Resource Allocation, and then the critic component evaluates the handoff and resource allocation actions of the actor component to guide the optimization of actions in the actor component. The simulation results show that HCRA can obtain better performance than benchmark algorithms.
- 29 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 10 (15) , 13370-13384
Handoff Control and Resource Allocation for RAN Slicing in IoT Based on DTN: An Improved Algorithm Based on Actor–Critic Framework
As a three-layer association of Internet of things equipment (IoTE) -network slicing (NS) -base station (BS) in radio access network (RAN) slicing, handoff control and resource allocation has become an important but complicated issue. In addition, the centralized controller has a difficult grasping the network situation in real time. In view of this, the problem of handoff control in the RAN slicing are investigated in the digital twin network (DTN), with the goal of maximizing the long-term utility about user satisfaction and handoff cost. Then an improved algorithm based on actor-critic framework is suggested, which is called HCRA. Specifically, the actor component contains neural networks for Handoff Control and an optimizer for Resource Allocation, and then the critic component evaluates the handoff and resource allocation actions of the actor component to guide the optimization of actions in the actor component. The simulation results show that HCRA can obtain better performance than benchmark algorithms.