Fang Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Fang Li has a total of 17 co-authors. Here are the top co-authors by number of shared publications:
Linsong Li
Guilin University of Electronic Technology
2 shared publications- Shuai Cao
Guilin University of Electronic Technology
2 shared publications Xiaonan Luo
2 shared publicationsJiacheng Ni
School of Computer Science and Information Security, Guilin University of Electronic Technology, Guilin, China
1 shared publication- Jiacheng Ni
Guilin University of Electronic Technology
1 shared publication Jicheng Yao
Guilin University of Electronic Technology
1 shared publication
Latest Publications
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (19) , 28658-28667
MSGFN: Multiscale Gated Fusion Network for UWB NLOS/LOS Classifications
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (19) , 28658-28667
MSGFN: Multiscale Gated Fusion Network for UWB NLOS/LOS Classifications
- 30 September 2026
- Published by Elsevier in Ocean Engineering
- Vol. 368
A safety-consistent hybrid global–local path planning framework for AUVs in bathymetry-constrained 3D marine environments
- 30 September 2026
- Published by Elsevier in Ocean Engineering
- Vol. 368
A safety-consistent hybrid global–local path planning framework for AUVs in bathymetry-constrained 3D marine environments
- Open Access
- 11 August 2026
- Published by Elsevier in SSRN Electronic Journal
RSPA-FKGC: Relation-conditioned Semantic Discrimination and Decoupled Path Adaptation for Few-shot Knowledge Graph Completion
Few-shot knowledge graph completion (FKGC) aims to predict missing entities for relations unseen during training from only a small set of support triples. Local topology provides a basis for cross-relation transfer, while entity descriptions, relation semantics, and multi-hop paths offer complementary information. These sources, however, differ substantially in reliability and ranking function. Uniformly fusing them may distort the structural metric space or leave auxiliary evidence too weak to correct top-ranked candidates. We propose RSPA-FKGC, a role-separated framework for coordinating structural, textual, and path evidence. It constructs entity-pair representations through relation-aware structural encoding and gated entity-text residuals, models task semantics and candidate compatibility through a relation-text prototype prior and a relation-conditioned TextScore, and learns a task-conditioned two-hop path adapter only after freezing the structural-semantic backbone. At inference, the normalized score gap between the two highest-ranked candidates from the structure-path branch adaptively controls the magnitude of textual correction. Under the 5-shot evaluation protocol used in this study, RSPA-FKGC achieves MRR values of 0.472 on FB15K237-One and 0.377 on Wiki-One. Incremental experiments show that relation-conditioned TextScore raises MRR on FB15K237-One from 0.430 to 0.456; auxiliary ranking and exponential moving averaging further produce a path-free backbone with 0.463 MRR, and decoupled path adaptation ultimately increases MRR to 0.472. These results indicate that assigning heterogeneous evidence distinct representation, optimization, and ranking responsibilities is more effective for few-shot link prediction than treating the evidence sources as interchangeable features.
- 11 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessRSPA-FKGC: Relation-conditioned Semantic Discrimination and Decoupled Path Adaptation for Few-shot Knowledge Graph Completion
Few-shot knowledge graph completion (FKGC) aims to predict missing entities for relations unseen during training from only a small set of support triples. Local topology provides a basis for cross-relation transfer, while entity descriptions, relation semantics, and multi-hop paths offer complementary information. These sources, however, differ substantially in reliability and ranking function. Uniformly fusing them may distort the structural metric space or leave auxiliary evidence too weak to correct top-ranked candidates. We propose RSPA-FKGC, a role-separated framework for coordinating structural, textual, and path evidence. It constructs entity-pair representations through relation-aware structural encoding and gated entity-text residuals, models task semantics and candidate compatibility through a relation-text prototype prior and a relation-conditioned TextScore, and learns a task-conditioned two-hop path adapter only after freezing the structural-semantic backbone. At inference, the normalized score gap between the two highest-ranked candidates from the structure-path branch adaptively controls the magnitude of textual correction. Under the 5-shot evaluation protocol used in this study, RSPA-FKGC achieves MRR values of 0.472 on FB15K237-One and 0.377 on Wiki-One. Incremental experiments show that relation-conditioned TextScore raises MRR on FB15K237-One from 0.430 to 0.456; auxiliary ranking and exponential moving averaging further produce a path-free backbone with 0.463 MRR, and decoupled path adaptation ultimately increases MRR to 0.472. These results indicate that assigning heterogeneous evidence distinct representation, optimization, and ranking responsibilities is more effective for few-shot link prediction than treating the evidence sources as interchangeable features.
- Open Access
- 29 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 134590-134600
RIM-Net: A Real-Imaginary-Magnitude Network for NLOS/LOS Identification in UWB Indoor Positioning Systems
In UWB-based indoor positioning research, NLOS signals severely degrade localization accuracy. To address the NLOS/LOS disRIMination challenge, this paper proposes the RIM-Net model, which enhances CIR signal feature learning by integrating residual blocks and LSTM architectures. Unlike conventional methods that process only the magnitude of CIR signals, RIM-Net explicitly models both real and imaginary components of CIR signals to improve classification accuracy. Experimental results demonstrate that RIM-Net achieves an average accuracy of 92.36% on benchmark datasets while maintaining generalization capabilities above 85% in unseen environments. Furthermore, the optimization of the TOA algorithm validated the effectiveness of RIM-Net in localization, achieving an average performance improvement of 9.66%. This approach provides an effective solution for NLOS/LOS identification in complex channel conditions within UWB positioning systems.
- 29 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 134590-134600
Open AccessRIM-Net: A Real-Imaginary-Magnitude Network for NLOS/LOS Identification in UWB Indoor Positioning Systems
In UWB-based indoor positioning research, NLOS signals severely degrade localization accuracy. To address the NLOS/LOS disRIMination challenge, this paper proposes the RIM-Net model, which enhances CIR signal feature learning by integrating residual blocks and LSTM architectures. Unlike conventional methods that process only the magnitude of CIR signals, RIM-Net explicitly models both real and imaginary components of CIR signals to improve classification accuracy. Experimental results demonstrate that RIM-Net achieves an average accuracy of 92.36% on benchmark datasets while maintaining generalization capabilities above 85% in unseen environments. Furthermore, the optimization of the TOA algorithm validated the effectiveness of RIM-Net in localization, achieving an average performance improvement of 9.66%. This approach provides an effective solution for NLOS/LOS identification in complex channel conditions within UWB positioning systems.
- Open Access
- 20 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 1760-1765
Penetration Test Path Discovery Based on NHSC-PPO
Aiming at the problems that Deep Q Network (DQN) and its variant Deep Reinforcement Learning (DRL) algorithm are not efficient and difficult to converge in the penetration test attack path discovery scenario, this study introduces the Normalized Entropic Dynamic Reward Scaled Clipped-Proximal Policy Optimization Algorithm (NHSC-PPO) for determining the optimal attack path. This algorithm enhances the Proximal Policy Optimization (PPO) algorithm, and introduces advantage function normalization, policy entropy, dynamic reward scaling, and gradient clipping techniques. We first convert the network topology and vulnerability information into an attack graph by MulVAL, and then train the Agent to continuously explore and update the strategy on the attack graph by using the NHSC-PPO algorithm, and finally find the attack path with the highest reward value in the dataset. Experimental results in various scenarios show that, compared with PPO, Double Dueling DQN and DQN algorithms, the penetration testing path discovery of NHSC-PPO algorithm can more efficiently and accurately converge to the path with the highest reward, and then determining the optimal attack path.
- 20 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 1760-1765
Open AccessPenetration Test Path Discovery Based on NHSC-PPO
Aiming at the problems that Deep Q Network (DQN) and its variant Deep Reinforcement Learning (DRL) algorithm are not efficient and difficult to converge in the penetration test attack path discovery scenario, this study introduces the Normalized Entropic Dynamic Reward Scaled Clipped-Proximal Policy Optimization Algorithm (NHSC-PPO) for determining the optimal attack path. This algorithm enhances the Proximal Policy Optimization (PPO) algorithm, and introduces advantage function normalization, policy entropy, dynamic reward scaling, and gradient clipping techniques. We first convert the network topology and vulnerability information into an attack graph by MulVAL, and then train the Agent to continuously explore and update the strategy on the attack graph by using the NHSC-PPO algorithm, and finally find the attack path with the highest reward value in the dataset. Experimental results in various scenarios show that, compared with PPO, Double Dueling DQN and DQN algorithms, the penetration testing path discovery of NHSC-PPO algorithm can more efficiently and accurately converge to the path with the highest reward, and then determining the optimal attack path.