CV

Undergraduate

• Expected graduation date 2025.7
• Double major in Mathematics and Physics + Industrial Engineering

Exchange

• Visited as an exchange student in the Faculty of Arts and Science.
• Joined the research group of Prof. Chi-Guhn Lee in the Department of Mechanical and Industrial Engineering.

A Machine Learning Enhanced Decomposition Approach to Solving Maximum Clique on Quantum Annealers

• We proposed an exact decomposition method to separate graphs for the maximum clique problem via vertex separators. In order to improve the performance of the decomposition approach, we developed a supervised learning method and a Monte-Carlo Tree Search method to improve the choice of the vertex. We further proposed a reinforcement learning method to identify vertex separators. Results showed that our method could significantly improve the performance of the heuristic method for decomposition.

Research on Large-Scale Mixed Integer Programs Based on Machine Learning Methods

• I proposed a hypergraph-based method to completely encode quadratically constrained quadratic programs as a hypergraph and improved existing hypergraph neural networks to predict optimal solutions. I proposed a neighborhood crossover algorithm based on McCormick Relaxation, which enabled small-scale solvers to solve large-scale problems and strengthened traditional large neighborhood search methods by paralleling.I led the project with two other undergraduates, developed all the code and written most part of our paper (which is currently under review for ICLR 2024). Results showed that our method outperforms Gurobi and SCIP significantly on large problems. For the next step, I am considering

• strengthen the theoretical analysis of the hypergraph-based method,
• use RL to improve the neighborhood selection process,
• select tight constraints to further improve performance.

Curriculum Learning Algorithm Development and Open-Source Framework Construction

• I Contributed to the curriculum learning benchmark repo and developed the code based on the RL-teacher paper. I conducted experiments for the RL-teacher algorithm by systematically testing all combinations of parameters. I co-authored a paper which is currently under review for CVPR 2024.

Online TradingPlatform

• This is course project of a online trading plagform written in SQL and Python.

A ML-enhanced Decomposition Approach to Solving the Maximum Clique Problem

• This is a course project for MIE1666 (Machine Learning for Mathematical Programming). Won Quantum Annealers.

• First Prize in the 31st China Physics Olympiad (Shanghai Region)

• First Prize in the Tsinghua University Physics Autumn Camp