Shi Zhuo Looi
California institute of technology
Mathematics (Division of physics, mathematics and astronomy)
Biography
Looi works in mathematical analysis, evolutionary PDEs, and AI. A core part of his research in analysis and PDEs involves the study of inequalities, including functional and polynomial inequalities, which are intimately related to many areas of pure and applied mathematics. In the AI-math domain, his contributions include proving theorems on the controllability of self-attention in “What’s the Magic Word?”, leading the data team at Project Numina (winner of the 2024 AIMO Prize), and advancing Lean-4 formalization in PDEs. He is a founding scientific advisor to ScienceStack, a platform for interactive, machine-readable papers. His work aims to merge machine learning, formal methods, and math to build reliable and reproducible reasoning workflows in mathematics.
SRP Project Title
AI-Driven Methods for Discovering and Proving Mathematical Inequalities
NAIRR Project
AI-Driven Methods for Discovering and Proving Mathematical Inequalities
Topical Areas
Applied Computer Science; Applied Mathematics; Artificial Intelligence and Intelligent Systems; Computer Science; High Performance Computing; Performance Evaluation and Benchmarking; Training
Abstract
Our proposed work advances AI for Accelerating Science and Discovery, a core focus area of the NAIRR Pilot, by developing a specialized, AI-driven framework for discovering and proving mathematical inequalities. Inequalities are foundational tools across scientific and engineering domains, from verifying stability in partial differential equations to bounding error terms in applied mathematics and physics. By combining large language models, reinforcement learning (RL), and symbolic computation, we aim to automate and accelerate the process of finding new results in both classical and cutting-edge mathematical research. This will facilitate the proof of inequalities and has the potential to enable new discoveries in science and engineering where precise mathematical bounds are important.
Desired Skills
Experience training models and/or handling datasets for language models
Lightning Talk Title
AI-Driven Discovery and Proof of Mathematical Inequalities