Scholar
Li Zeng
Google Scholar ID: AH1tK9IAAAAJ
Peking University
Citations & Impact
Citations
187
H-index
7
i10-index
5
Publications
15
Co-authors
8
list available
Contact
Publications
Resume (English only)
Academic Achievements
- Published multiple high-impact papers, including in CCF A venues: IEEE TKDE and ICDE
- “WindGP: Efficient Graph Partitioning on Heterogenous Machines” (arXiv, 2024)
- Second author of “LocMoE: A Low-Overhead MoE for Large Language Model Training” (IJCAI, 2024)
- Multiple papers received GraphChallenge Innovation Awards (e.g., RaftGP, HTC)
- Published in DEXA, HPCC, DASFAA, WWW, FCS, NLPCC on topics of graph computing and knowledge graph querying
- Co-authored Chinese paper “Regular Path Queries on Large Graph Data” in Acta Scientiarum Naturalium Universitatis Pekinensis (2018)
Research Experience
- Primary developer of the graph database system gStore, contributing over 5 million lines of code and improving performance by 100× and scalability by 40×
- Led a team of more than 10 members, assigning and qualifying them for respective modules
- Redesigned the data loading architecture for hybrid node/edge features in large-scale graph ML systems, achieving >2× speedup
- Developed a memory clipping module that enables user-defined sampling during data loading under memory constraints, reducing memory usage by 31% with only 1% model performance loss
- Conducted surveys and proposed four general optimization techniques for subgraph isomorphism on CPU
- Designed novel GPU-friendly data structures and join algorithms for subgraph isomorphism, achieving >10× speedup
- Optimized GPU implementations of shortest path and triangle counting algorithms, both achieving >2× speedup
