CV
Education, research experience, and publications.
Contact Information
| Name | Guanghua Li |
| Professional Title | Ph.D. Student in Data Science and Analytics |
| gli945@connect.hkust-gz.edu.cn | |
| Location | Guangzhou, China |
Professional Summary
Ph.D. student at HKUST (Guangzhou), advised by Prof. Qiong Luo, with expected graduation in January 2027. Research focuses on database systems, GPU acceleration, graph query processing, and multi-vector retrieval, with a long-term interest in data systems for the AI era.
Experience
-
2021 - 2022 Hong Kong, China
Research Assistant
The Chinese University of Hong Kong
Database Research Group, Department of Systems Engineering and Engineering Management.
- Advisor: Prof. Jeffrey Xu Yu
Education
-
2022 - present Guangzhou, China
Ph.D. (in progress)
The Hong Kong University of Science and Technology (Guangzhou)
Data Science and Analytics
- Advisor: Prof. Qiong Luo
- Expected graduation: January 2027
-
2017 - 2022 Wuhan, China
Bachelor's degree
Wuhan University
Computer Science and Technology
- Advisor: Prof. Yuanyuan Zhu
Publications
-
2027 Vora: A Vector-Based Engine for Scalable GPU-Accelerated Subgraph Query Processing
ICDE 2027 (accepted)
Guanghua Li, Hao Zhang, and Qiong Luo. A single-GPU engine with fused vector operations and query decomposition for subgraph processing beyond device memory.
-
2026 Efficient GPU-Based Continuous Subgraph Matching on Batch Updates
APWeb 2026
Guanghua Li and Xibo Sun (equal contribution), Qiong Luo, and Lijun Chang. Query-oriented batch processing avoids duplicate matches; GCSM-BU implements the approach on GPUs.
-
2024 TenGraph: A Tensor-Based Graph Query Engine
Proceedings of the VLDB Endowment, 17(13): 4571–4584
Guanghua Li, Hao Zhang, Xibo Sun, Qiong Luo, and Yuanyuan Zhu. A PyTorch-based graph query engine that runs on multicore CPUs and GPUs using a common tensor representation.
-
2023 Community Search: A Meta-Learning Approach
ICDE 2023, pp. 2358–2371
Shuheng Fang, Kangfei Zhao, Guanghua Li, and Jeffrey Xu Yu. Conditional Graph Neural Process (CGNP) learns shared knowledge across community-search tasks and adapts to new tasks with limited labeled data.
-
2021 FANE: A Fusion-Based Attributed Network Embedding Framework
APWeb-WAIM 2021, LNCS 12858, pp. 53–60
Guanghua Li, Qiyan Li, Jingqiao Liu, Yuanyuan Zhu, and Ming Zhong. A framework that learns separate embeddings for network structure and attributes, then fuses them for downstream tasks such as node classification and link prediction.
Projects
-
VectorHit
Ongoing research on GPU-accelerated multi-vector retrieval.
-
Vora
Scalable GPU-accelerated subgraph query processing.
- Developed VectorFlux in C++ on Thrust and RAPIDS Memory Manager, with heterogeneous memory support, non-owning views, and lazy operator fusion.
- Implemented expansion, semi-join, and anti-join operators that produce columnar matching results.
- Decomposed queries into tasks with bounded input-shard memory, using CPU–GPU transfer volume as the optimization objective.
- The accepted paper reports up to 12× speedup over optimized CPU systems and up to 7× over the best GPU baseline on the evaluated workloads.
-
TenGraph
A tensor-based graph query engine built on PyTorch.
- Represented graph structure, properties, and intermediate results with one-dimensional tensors.
- Implemented matching, filtering, projection, aggregation, and ordering through tensor operations.
- Used one codebase for multicore CPU and GPU execution; the paper reports 50–100× GPU-versus-CPU speedups on evaluated workloads.
-
GCSM-BU
Continuous subgraph matching on batch updates.
- Co-developed QO-CSM, a query-oriented formulation that avoids duplicate matches without explicit duplicate removal.
- Combined two-level indexing, dynamic breadth-first/depth-first enumeration, and backtracking flattening to balance GPU parallelism, memory use, and workload.
Awards
-
2018 National Scholarship for Undergraduate Students
Ministry of Education of China