About Me
I am a fourth-year Ph.D. student at Purdue University in the Davidson School of Chemical Engineering (Li Group).
I received M.Phil. from the Department of Chemical Engineering and Biotechnology at University of Cambridge (2021), and B.Eng. from the Department of Chemical and Environmental Engineering at University of Nottingham (2020).
I am actively seeking opportunities for internships and research collaborations.
My research interests lie broadly at the interface between machine learning and optimization, with a focus on accelerating hard decision-making problems, improving explainability of optimization models, and preserving feasibility in deep learning models. Current work follows three connected directions.
Learning to Optimize
Many industrial decision-making problems can be formulated as combinatorial optimization problems, yet solving them to global optimality remains computationally challenging. My research develops learning-based approaches to accelerate the branch-and-bound methods without sacrificing correctness guarantees.
Related publication: Solving Max-Cut to Global Optimality via Feasibility-Preserving Graph Neural Networks
LLMs for Optimization
Optimization models are often developed by experts but used by domain practitioners who need to understand formulations, diagnose infeasibility, and evaluate possible model changes. This direction proposes LLM-based systems that connect optimization models, solvers, and users through natural language.
Related publications: OptiChat: Bridging Optimization Models and Practitioners with Large Language Models; Diagnosing Infeasible Optimization Problems Using Large Language Models
Deep Learning Models with Hard Constraints
Deep learning models are widely used in science and engineering as surrogates, but their predictions may violate physical laws, domain knowledge and operational requirements. This work explores neural architectures to enforce input-dependent hard constraints with minimal computational overhead.
Related publications: Enforcing Hard Linear Constraints in Deep Learning Models with Decision Rules; Physics-informed neural networks with hard linear equality constraints