CV

Education

• B.S. in Computer Science and Technology, Southern University of Science and Technology, expected 2023
• GPA: 3.89/4.00, Rank: 3/157
• Key Courses: Probability and Statistics, Data Structure and Algorithm, Database Principles, Computer Architecture, Information Theory, Data Mining, Machine Learning

Research Interest

System and algorithms for large-scale data

PUBLICATIONS

Manu: A Cloud Native Vector Database Management System

R. Guo, X. Luan, L. Xiang, X. Yan, X. Yi, J. Luo, Q. Cheng, W. Xu, J. Luo, F. Liu, et al., Manu: A Cloud Native Vector Database Management System, The International Conference on Very Large Databases 2022.

Research Experience

• Research Topic: Learning Methods for Database Tuning

  Role: Research Assistant, Supervisor: Prof. Guoliang Li in Tsinghua Database Group

  Database has hundreds of knobs that significantly influence the performance of a database. In recent years, a number of learning based methods have been proposed to tune the knobs automatically.

  • Investigated and implemented database tuning methods, including OtterTune, CDBTune, and QTune, etc.
  • Explore the possibility of transferring tuning models from one tuning environment to another, in order to reduce the tuning time for database users.

• SIGMOD’22 Programming Contest on Entity Blocking, Finalist

  Role: Leader, Supervisor: Prof. Bo Tang and Prof. Xiao Yan in Database Group

  Entity blocking, a step for entity matching task, is used to quickly filter out the non-matches. The contest prepared million-scale datasets from e-commercial websites, and participants were required to design a system that generates small candidate sets of pairs with high recall. Our solution ended up with a finalist award.

  • Led a team of three students, proposed to use pre‑trained neural network model to transform entity description sentences into representative vectors.
  • Adopted a HNSW index, increased graph connectivity in HNSW to improve performance and limited the size of priority queue in the searching process to reduce search time, as a result the similar vectors were matched quickly with high probability.

• Research Topic: Graph Summarization

  Role: Leader, Supervisor: Prof. Bo Tang and Prof. Xiao Yan in Database Group

  Given a graph, graph summarization represents it as a summary graph and edge corrections, with the goal of minimizing the overall size of the two parts. As a result, graph summarization could be used for compression and pattern mining. I proposed a new solution that improves the compression rate and execution time.

  • Investigated graph summarization problem, tested related works, i.e., Randomized, SWeG, LDME.
  • Assessed different hashing schemes, including MinHash, DOPH and Asymmetric LSH.
  • Explored node encoding and maximum inner product search to make improvements of compression rate.
  • Designed a LSH tree structure after analyzing the problem in detail to solve overfitting problem. Proposed an algorithm based on the LSH tree structure to summarize graph more efficiently.
  • Conducted experiments and proofed the superiority of the algorithm on compression rate and elapsed time.

• NeurIPS’21 Billion‑Scale Vector Search Competition (Track 2), Champion

  Role: Team Member, Supervisor: Prof. Bo Tang and Prof. Xiao Yan in Database Group

  Track 2 conducts vector search on 6 billion-scale datasets with 64GB DRAM and 1TB SSD, and compares the recall obtained at 1500 query/s. The key is to optimize data read for the low bandwith and long latency of SSD. Our proposal boosts the recall competition baseline on range search by more than 70%.

  • Investigated, implemented and assessed key vector search methods, including Locality Sensitive Hashing (LSH), Proximity Graph, Vector Quantization, etc.
  • Investigated the characteristics of SSD I/O in detail to make fully use of SSD.
  • Proposed bucket graph solution for billion-scale datasets that clusters vectors into 4KB buckets to align with SSD blocks for efficient read. Built HNSW to navigate among the bucket centers in DRAM.
  • Implemented scalar quantization for read reduction, HNSW for bucket center search and Python‑C++ interfaces that connected our implementation with the competition test environment.
  • Conducted experiments to verify the superiority of our solution, which boosts the recall of competition baseline on range search by more than 70%.
  • Applied our solution to Milvus, a cloud-native vector database system, which has been adopted by many enterprises including Tencent, ebay and NVIDIA.
  • Published a paper that describes our solution on VLDB 2022 conference about the system.

Teaching Experience

Southern University of Science and Technology

Role: Teaching Assistant

• Machine Learning, Fall 2022 by Professor Qi Hao
  • Created and maintained an online judge system
  • Helped international students in the lab and after lectures
• Discrete Math, Spring 2021 by Professor Adam Ghandar
  • Helped develop assignments
  • Answered questions after lectures
  • Graded assignments and midterm exams

HONORS & AWARDS

• Chinese National Scholarship, 2022
• Finalist, SIGMOD’22 Programming Contest on Entity Blocking, 2022
• Honorable Mention, The International Mathematical Contest in Modeling, 2022
• Champion, NeurIPS’21 Billion-Scale Vector Search Competition, 2021
• 3rd Prize, University Merit Student Scholarship, 2021
• 2nd Prize, College ACM Contest, 2020
• 2nd Prize, University Merit Student Scholarship, 2020

Skills & Tools

• Programming Languages: Java, C/C++, Python, SQL, Matlab, Verilog HDL
• English Proficiency: IELTS (7.5), TOEFL iBT (105), GRE General Test (325)
• Tools: Jetbrains (IntelliJ IDEA, Clion, PyCharm, DataGrip), VSCode, Jupyter Notebook, Unity, Ubuntu