Event
Ph.D. Research Proposal: Mingju Liu
Tuesday, February 3, 2026
1:30 p.m.
AVW 1146
ANNOUNCEMENT: Ph.D. Research Proposal Exam
Name: Mingju Liu
Committee:
Professor Cunxi Yu (Chair)
Professor Kaiqing Zhang
Professor Gang Qu
Date/time: Tuesday, February 3, 2026 at 1:30pm
Location: AVW 1146
Title: Sampling-Opt: Bridging Differentiable Learning and Combinatorial Optimization for Next-Generation EDA
Abstract:
Combinatorial optimization lies at the heart of Electronic Design Automation (EDA), spanning tasks from high-level scheduling to logic synthesis and verification. Traditional methods face a critical trade-off: exact solvers (such as ILP and SAT solvers) guarantee optimality but suffer from poor scalability due to exponential time complexity, while heuristic-based methods offer speed at the cost of sub-optimal results.
My research proposes a paradigm called "Sampling-Opt" to advance the resolution of classic combinatorial optimization problems in EDA. This approach leverages various sampling-based techniques—specifically the Differentiable Gumbel-Softmax trick and Reinforcement Learning (RL)—to navigate massive search spaces in specific fields efficiently. By treating discrete optimization decisions as differentiable sampling steps, we can rapidly identify high-quality solutions or significantly reduced search spaces.
Furthermore, I propose a hybrid methodology that employs traditional solvers to refine the solutions obtained from sampling-based methods. This approach aims to combine the scalability of learning-based sampling with the optimality of classic exact solvers. Beyond theoretical contributions, I will demonstrate real-world applications capable of being extended to industrial-level tools.
