Yaqiong Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yaqiong Liu has a total of 74 co-authors. Here are the top co-authors by number of shared publications:
- Guochu Shou
Beijing Key Laboratory of Network System Architecture and Convergence, Beijing University of Posts and Telecommunications, Beijing, China
12 shared publications - Mugen Peng
Beijing University of Posts and Telecommunications
6 shared publications - Yihong Hu
Beijing University of Posts and Telecommunications, Beijing, China
6 shared publications Guochu Shou
Beijing University of Posts and Telecommunications, Beijing, China
5 shared publicationsHock Soon Seah
Nanyang Technol Univ, Sch Comp Sci & Engn, Singapore 639798, Singapore
5 shared publicationsYihong Hu
Beijing Key Laboratory of Network System Architecture and Convergence, Beijing University of Posts and Telecommunications, Beijing, China
4 shared publications
Latest Publications
- Open Access
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 6421-6432
FlightGen: Controllable Flight Trajectory Generation with Physics-Constrained Diffusion
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 6421-6432
Open AccessFlightGen: Controllable Flight Trajectory Generation with Physics-Constrained Diffusion
- 6 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 23, 3026-3038
Entity-Level Autoregressive Relational Triple Extraction Toward Knowledge Graph Construction for Network Operation and Maintenance
With the significant increase of communication network scales, intelligent Network Operation and Maintenance (NOM) becomes essential. Knowledge Graphs (KGs) are a key enabler for intelligent NOM, and Relational Triple Extraction (RTE) plays a critical role in KG construction. However, most existing RTE researches rely on general-domain corpora, with limited exploration into the specialized domain. In this paper, we identify a novel challenge in Chinese NOM corpus — Segmented Entity, which has garnered little attention in prior works. To address it, this paper proposes an Entity-level Autoregressive RTE (EARTE) method, which incorporates an innovative Segmented-BIO (Begin, Inside, Outside) tagging scheme. Furthermore, we construct the CMIM23-NOM1-RA, the first high-quality restricted domain RTE dataset for NOM. Throughout the experimentation, we meticulously reproduce all baselines and provide a comprehensive analysis. The results show that EARTE achieves the best performance on CMIM23-NOM1-RA. EARTE’s F1 scores surpass those of the best-performing baselines by 0.4%, 2.7%, and 0.8% under the strict criterion, the lenient criterion, and the setting focusing only on segmented entities, respectively. Finally, our codes, dataset, and reproduction guidelines are publicly available at: https://github.com/JYzzzzzz/PEAR-RTE.
- 6 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 23, 3026-3038
Entity-Level Autoregressive Relational Triple Extraction Toward Knowledge Graph Construction for Network Operation and Maintenance
With the significant increase of communication network scales, intelligent Network Operation and Maintenance (NOM) becomes essential. Knowledge Graphs (KGs) are a key enabler for intelligent NOM, and Relational Triple Extraction (RTE) plays a critical role in KG construction. However, most existing RTE researches rely on general-domain corpora, with limited exploration into the specialized domain. In this paper, we identify a novel challenge in Chinese NOM corpus — Segmented Entity, which has garnered little attention in prior works. To address it, this paper proposes an Entity-level Autoregressive RTE (EARTE) method, which incorporates an innovative Segmented-BIO (Begin, Inside, Outside) tagging scheme. Furthermore, we construct the CMIM23-NOM1-RA, the first high-quality restricted domain RTE dataset for NOM. Throughout the experimentation, we meticulously reproduce all baselines and provide a comprehensive analysis. The results show that EARTE achieves the best performance on CMIM23-NOM1-RA. EARTE’s F1 scores surpass those of the best-performing baselines by 0.4%, 2.7%, and 0.8% under the strict criterion, the lenient criterion, and the setting focusing only on segmented entities, respectively. Finally, our codes, dataset, and reproduction guidelines are publicly available at: https://github.com/JYzzzzzz/PEAR-RTE.
- Open Access
Accurate Network Traffic Matrix Prediction via LEAD: a Large Language Model-Enhanced Adapter-Based Conditional Diffusion Model
Driven by the evolution toward 6G and AI-native edge intelligence, network operations increasingly require predictive and risk-aware adaptation under stringent computation and latency constraints. Network Traffic Matrix (TM), which characterizes flow volumes between nodes, is a fundamental signal for proactive traffic engineering. However, accurate TM forecasting remains challenging due to the stochastic, non-linear, and bursty nature of network dynamics. Existing discriminative models often suffer from over-smoothing and provide limited uncertainty awareness, leading to poor fidelity under extreme bursts. To address these limitations, we propose LEAD, a Large Language Model (LLM)-Enhanced Adapter-based conditional Diffusion model. First, LEAD adopts a "Traffic-to-Image" paradigm to transform traffic matrices into RGB images, enabling global dependency modeling via vision backbones. Then, we design a "Frozen LLM with Trainable Adapter" model, which efficiently captures temporal semantics with limited computational cost. Moreover, we propose a Dual-Conditioning Strategy to precisely guide a diffusion model to generate complex, dynamic network traffic matrices. Experiments on the Abilene and GEANT datasets demonstrate that LEAD outperforms all baselines. On the Abilene dataset, LEAD attains a remarkable 45.2% reduction in RMSE against the best baseline, with the error margin rising only marginally from 0.1098 at one-step to 0.1134 at 20-step predictions. Meanwhile, on the GEANT dataset, LEAD achieves a 0.0258 RMSE at 20-step prediction horizon which is 27.3% lower than the best baseline.
Open AccessAccurate Network Traffic Matrix Prediction via LEAD: a Large Language Model-Enhanced Adapter-Based Conditional Diffusion Model
Driven by the evolution toward 6G and AI-native edge intelligence, network operations increasingly require predictive and risk-aware adaptation under stringent computation and latency constraints. Network Traffic Matrix (TM), which characterizes flow volumes between nodes, is a fundamental signal for proactive traffic engineering. However, accurate TM forecasting remains challenging due to the stochastic, non-linear, and bursty nature of network dynamics. Existing discriminative models often suffer from over-smoothing and provide limited uncertainty awareness, leading to poor fidelity under extreme bursts. To address these limitations, we propose LEAD, a Large Language Model (LLM)-Enhanced Adapter-based conditional Diffusion model. First, LEAD adopts a "Traffic-to-Image" paradigm to transform traffic matrices into RGB images, enabling global dependency modeling via vision backbones. Then, we design a "Frozen LLM with Trainable Adapter" model, which efficiently captures temporal semantics with limited computational cost. Moreover, we propose a Dual-Conditioning Strategy to precisely guide a diffusion model to generate complex, dynamic network traffic matrices. Experiments on the Abilene and GEANT datasets demonstrate that LEAD outperforms all baselines. On the Abilene dataset, LEAD attains a remarkable 45.2% reduction in RMSE against the best baseline, with the error margin rising only marginally from 0.1098 at one-step to 0.1134 at 20-step predictions. Meanwhile, on the GEANT dataset, LEAD achieves a 0.0258 RMSE at 20-step prediction horizon which is 27.3% lower than the best baseline.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. PP  (00189545) , 1-15
Multi-Agent Deep Reinforcement Learning for Joint Optimization of Latency and Energy Consumption Oriented to Communication-Sensing-Computation Integration in IoV
The rapid development of the Internet of Vehicles (IoV) has driven the need for efficient Integration of Communication, Sensing, and Computation (ICSC) to support advanced applications such as autonomous driving and real-time traffic management. However, the dynamic and resource-constrained nature of IoV environment poses significant challenges in achieving optimal performance, particularly in balancing multiple optimization objectives such as latency and energy consumption. Traditional optimization methods, such as Genetic Algorithms (GA), often struggle with inefficiency and unpredictability in dynamic scenarios. In this paper, we first propose an IoV Communication-Sensing-Computation Integration architecture to address the problem that the traditional IoV architectures cannot meet the needs of the ICSC. Then, we formulate the joint optimization of latency and energy consumption in ICSC as a constrained Multi-objective Optimization Problem (MOP). We employ Multi-Agent Deep Reinforcement Learning (MADRL) with a transformer module, combined with a weighted decomposition method and a neighborhood parameter transfer technique, to solve the problem. Simulation results show that our approach outperforms traditional methods in terms of convergence and diversity, and has a lower running time.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. PP  (00189545) , 1-15
Multi-Agent Deep Reinforcement Learning for Joint Optimization of Latency and Energy Consumption Oriented to Communication-Sensing-Computation Integration in IoV
The rapid development of the Internet of Vehicles (IoV) has driven the need for efficient Integration of Communication, Sensing, and Computation (ICSC) to support advanced applications such as autonomous driving and real-time traffic management. However, the dynamic and resource-constrained nature of IoV environment poses significant challenges in achieving optimal performance, particularly in balancing multiple optimization objectives such as latency and energy consumption. Traditional optimization methods, such as Genetic Algorithms (GA), often struggle with inefficiency and unpredictability in dynamic scenarios. In this paper, we first propose an IoV Communication-Sensing-Computation Integration architecture to address the problem that the traditional IoV architectures cannot meet the needs of the ICSC. Then, we formulate the joint optimization of latency and energy consumption in ICSC as a constrained Multi-objective Optimization Problem (MOP). We employ Multi-Agent Deep Reinforcement Learning (MADRL) with a transformer module, combined with a weighted decomposition method and a neighborhood parameter transfer technique, to solve the problem. Simulation results show that our approach outperforms traditional methods in terms of convergence and diversity, and has a lower running time.
- 12 June 2025
- Published by Springer Nature
- p. 255-267
The Management of Mobile Elements Based on Federal Learning and Knowledge Graph in Space-Air-Ground Integrated Network
- 12 June 2025
- Published by Springer Nature
- p. 255-267
The Management of Mobile Elements Based on Federal Learning and Knowledge Graph in Space-Air-Ground Integrated Network
- 4 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 22  (3) , 2546-2563
An Enhanced Reconfiguration for Deterministic Transmission in Time-Sensitive Networks
Time-aware shaper (TAS) is key to enabling deterministic guarantees in time-sensitive networks (TSN), but it requires precise configuration for specific traffic scenarios. Dynamic traffic scenarios are increasingly commonplace with the rise of emerging applications, necessitating TAS reconfiguration to adapt to the changes in traffic. However, existing mechanisms primarily reconfigure TAS by generating a new gate control list (GCL) and transitioning to it, which may lead to temporary violations of bounds on delay or jitter, providing no persistently deterministic guarantees. In this paper, we propose a novel TAS reconfiguration mechanism with the virtual GCL (VGCL) to satisfy the demands of dynamic traffic while guaranteeing deterministic transmission. It implements TAS reconfiguration for dynamic traffic by embedding different VGCLs into the GCL, avoiding the need for the GCL transition. Thus, the reconfiguration problem is modeled as an embedding problem by using the VGCL and we develop algorithms to solve it. Experimental results demonstrate that our mechanism can well reconfigure TAS for dynamic traffic without the GCL transition, and increase the reconfiguration success rate in various scenarios compared with the existing approaches.
- 4 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 22  (3) , 2546-2563
An Enhanced Reconfiguration for Deterministic Transmission in Time-Sensitive Networks
Time-aware shaper (TAS) is key to enabling deterministic guarantees in time-sensitive networks (TSN), but it requires precise configuration for specific traffic scenarios. Dynamic traffic scenarios are increasingly commonplace with the rise of emerging applications, necessitating TAS reconfiguration to adapt to the changes in traffic. However, existing mechanisms primarily reconfigure TAS by generating a new gate control list (GCL) and transitioning to it, which may lead to temporary violations of bounds on delay or jitter, providing no persistently deterministic guarantees. In this paper, we propose a novel TAS reconfiguration mechanism with the virtual GCL (VGCL) to satisfy the demands of dynamic traffic while guaranteeing deterministic transmission. It implements TAS reconfiguration for dynamic traffic by embedding different VGCLs into the GCL, avoiding the need for the GCL transition. Thus, the reconfiguration problem is modeled as an embedding problem by using the VGCL and we develop algorithms to solve it. Experimental results demonstrate that our mechanism can well reconfigure TAS for dynamic traffic without the GCL transition, and increase the reconfiguration success rate in various scenarios compared with the existing approaches.
- 21 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 25  (12) , 19457-19468
Joint Optimization of Latency and Energy Consumption via Deep Reinforcement Learning for Proximity Detection in Road Networks
The development of automatic driving and assisted driving breeds the problem of proximity detection in road networks, which plays a significant role in ensuring safe driving. Due to the fact that it is a time-sensitive task, the problem of proximity detection requires to judge whether two vehicles are close to each other in a very short time. However, the battery life and computation capacity of vehicles are limited in the actual scenario. Therefore, how to solve this problem with low latency and energy consumption is an important issue. In this paper, we investigate the Joint Optimization of the Latency and Energy consumption problem in the scenario of Proximity Detection, namely, JOLE-PD, which is formulated into a constrained multiobjective optimization problem. The DDPG-CMOA method is proposed to find a tradeoff between latency and energy consumption, achieving the Pareto optimal solutions. Besides, NSGA-II (Non-dominated Sorting Genetic Algorithm-II) and MOEA-D (Multi-objective Evolutionary Algorithm Based on Decomposition), as the typical algorithms to solve multiobjective optimization problem, are used as the baseline methods to compare the performance of DDPG-CMOA method under different parameters. The experimental results show the proposed DDPG-CMOA method requires much lower running time and has strong generalization ability. Moreover, the solutions obtained from the DDPG-CMOA method have a slightly better ability of convergence and diversity.
- 21 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 25  (12) , 19457-19468
Joint Optimization of Latency and Energy Consumption via Deep Reinforcement Learning for Proximity Detection in Road Networks
The development of automatic driving and assisted driving breeds the problem of proximity detection in road networks, which plays a significant role in ensuring safe driving. Due to the fact that it is a time-sensitive task, the problem of proximity detection requires to judge whether two vehicles are close to each other in a very short time. However, the battery life and computation capacity of vehicles are limited in the actual scenario. Therefore, how to solve this problem with low latency and energy consumption is an important issue. In this paper, we investigate the Joint Optimization of the Latency and Energy consumption problem in the scenario of Proximity Detection, namely, JOLE-PD, which is formulated into a constrained multiobjective optimization problem. The DDPG-CMOA method is proposed to find a tradeoff between latency and energy consumption, achieving the Pareto optimal solutions. Besides, NSGA-II (Non-dominated Sorting Genetic Algorithm-II) and MOEA-D (Multi-objective Evolutionary Algorithm Based on Decomposition), as the typical algorithms to solve multiobjective optimization problem, are used as the baseline methods to compare the performance of DDPG-CMOA method under different parameters. The experimental results show the proposed DDPG-CMOA method requires much lower running time and has strong generalization ability. Moreover, the solutions obtained from the DDPG-CMOA method have a slightly better ability of convergence and diversity.
- Open Access
- 21 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 1607-1616
A Knowledge-Enhanced Transformer-FL Method for Fault Root Cause Localization
Root cause analysis for faults is one of the core tasks in the operation and maintenance of communication networks. Although artificial intelligence techniques can be used to assist manual inspections, fault diagnosis is still a tough problem. In the scenario of fault root cause localization, insufficient associated information makes it difficult to accurately determine the root cause. Meanwhile, it is a big challenge to extract as many feature details as possible from limited information and then fully utilize them. Therefore, this paper proposes a Knowledge-Enhanced Transformer-FL method, namely, KETrans-FL, to address the problem of root cause localization, by treating it as a multi-class classification problem. Our method first constructs a knowledge graph for knowledge enhancement, which consists of four types of nodes (base station, alarm, fault and alarm level) and their relationships based on the network operations. This knowledge enhancement technique incorporates real operation data and other external knowledge (e.g., alarm level) to serve as the source of feature information and thus can extract statistical and embedded features. Then, our method designs a Trans-FL (Transformer-Focal Loss) model, which uses the adapted Transformer encoder to learn the correlation information between input features to generate classification probabilities and employs Focal Loss as the loss function to mitigate severe class imbalance in the multi-class classification problem. Experimental results show that our proposed KETrans-FL method achieves a classification accuracy of nearly 91% and an average AUC score of 93%, indicating a significant improvement on fault root cause localization compared to baseline models. In addition, experimental results also validate the remarkable effect of our knowledge enhancement technique on improving the final classification accuracy.
- 21 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 1607-1616
Open AccessA Knowledge-Enhanced Transformer-FL Method for Fault Root Cause Localization
Root cause analysis for faults is one of the core tasks in the operation and maintenance of communication networks. Although artificial intelligence techniques can be used to assist manual inspections, fault diagnosis is still a tough problem. In the scenario of fault root cause localization, insufficient associated information makes it difficult to accurately determine the root cause. Meanwhile, it is a big challenge to extract as many feature details as possible from limited information and then fully utilize them. Therefore, this paper proposes a Knowledge-Enhanced Transformer-FL method, namely, KETrans-FL, to address the problem of root cause localization, by treating it as a multi-class classification problem. Our method first constructs a knowledge graph for knowledge enhancement, which consists of four types of nodes (base station, alarm, fault and alarm level) and their relationships based on the network operations. This knowledge enhancement technique incorporates real operation data and other external knowledge (e.g., alarm level) to serve as the source of feature information and thus can extract statistical and embedded features. Then, our method designs a Trans-FL (Transformer-Focal Loss) model, which uses the adapted Transformer encoder to learn the correlation information between input features to generate classification probabilities and employs Focal Loss as the loss function to mitigate severe class imbalance in the multi-class classification problem. Experimental results show that our proposed KETrans-FL method achieves a classification accuracy of nearly 91% and an average AUC score of 93%, indicating a significant improvement on fault root cause localization compared to baseline models. In addition, experimental results also validate the remarkable effect of our knowledge enhancement technique on improving the final classification accuracy.
- 20 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 39  (1) , 220-226
Network Planning for Time-Sensitive Communications
Network planning is a critical process in the construction of communication networks. Time-sensitive communications require networks with time-aware scheduling and transmission mechanisms to guarantee the deterministic requirements of services, which is difficult to be satisfied by traditional network planning. This article proposes an SDN-enabled network planning solution for time-sensitive communications. Specifically, the solution introduces a layered network architecture by decoupling network planes with different functions, designs a planning strategy for deterministic services, and sorts out a time-sensitive network planning procedure that integrates topology construction, routing, and traffic scheduling. This article uses the scalefree topology construction method to construct delay-bounded network topology. Experimental results show that the solution can guarantee the bounded delay, improve the schedulability, and reduce the algorithm runtime. In addition, the solution is able to utilize network resources according to the capacity of each node.
- 20 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 39  (1) , 220-226
Network Planning for Time-Sensitive Communications
Network planning is a critical process in the construction of communication networks. Time-sensitive communications require networks with time-aware scheduling and transmission mechanisms to guarantee the deterministic requirements of services, which is difficult to be satisfied by traditional network planning. This article proposes an SDN-enabled network planning solution for time-sensitive communications. Specifically, the solution introduces a layered network architecture by decoupling network planes with different functions, designs a planning strategy for deterministic services, and sorts out a time-sensitive network planning procedure that integrates topology construction, routing, and traffic scheduling. This article uses the scalefree topology construction method to construct delay-bounded network topology. Experimental results show that the solution can guarantee the bounded delay, improve the schedulability, and reduce the algorithm runtime. In addition, the solution is able to utilize network resources according to the capacity of each node.
- Open Access
- 27 March 2024
- Published by MDPI AG in Electronics
- Vol. 13  (7) , 1246
Software-Defined Time-Sensitive Networking for Cross-Domain Deterministic Transmission
With the rapid development of Industrial 4.0, massive emerging time-critical applications pose new demands for industrial networks, such as strict latency boundaries, ultra-reliable transmission, and so on. Time-Sensitive Networking (TSN) is well suited for these demanding scenarios due to its support for deterministic transmission based on Ethernet. However, many use cases in real industrial scenarios involve non-TSN domains, and, thus, the network requires providing deterministic transmission across non-TSN domains to support them. To this end, this paper proposes a novel Software-Defined Time-Sensitive Networking (SD-TSN) framework that integrates the determinism of TSN and scalability of Software-Defined Networking (SDN). SD-TSN exploits the coordination between the coordinated controller and different domain controllers to bound non-deterministic queuing delays in a way that guarantees determinism. In particular, we designed a multi-domain time-aware traffic scheduling model, which harmonizes the time-aware shaper (TAS) in TSN and time-slots in non-TSN domains based on a global view, generating transmission schedules for each domain. A prototype testbed was built to conduct the experiments. The evaluation results demonstrate that the proposed method can efficiently achieve bounded delay and low delay variations in the transmission across non-TSN domains.
- 27 March 2024
- Published by MDPI AG in Electronics
- Vol. 13  (7) , 1246
Open AccessSoftware-Defined Time-Sensitive Networking for Cross-Domain Deterministic Transmission
With the rapid development of Industrial 4.0, massive emerging time-critical applications pose new demands for industrial networks, such as strict latency boundaries, ultra-reliable transmission, and so on. Time-Sensitive Networking (TSN) is well suited for these demanding scenarios due to its support for deterministic transmission based on Ethernet. However, many use cases in real industrial scenarios involve non-TSN domains, and, thus, the network requires providing deterministic transmission across non-TSN domains to support them. To this end, this paper proposes a novel Software-Defined Time-Sensitive Networking (SD-TSN) framework that integrates the determinism of TSN and scalability of Software-Defined Networking (SDN). SD-TSN exploits the coordination between the coordinated controller and different domain controllers to bound non-deterministic queuing delays in a way that guarantees determinism. In particular, we designed a multi-domain time-aware traffic scheduling model, which harmonizes the time-aware shaper (TAS) in TSN and time-slots in non-TSN domains based on a global view, generating transmission schedules for each domain. A prototype testbed was built to conduct the experiments. The evaluation results demonstrate that the proposed method can efficiently achieve bounded delay and low delay variations in the transmission across non-TSN domains.