Lab
Structure-Aware Learning and Modeling Lab
結構導向學習與建模實驗室（別名：格物致知實驗室）
Uncover the structure of things to expand the bounds of insight.
Structure-Aware Learning and Modeling Lab leverages the inherent structure of data and problems to build AI methods with rigorous mathematical foundations. We bridge mathematical theory, computational intuition, and implementable methods to advance research in geometric deep learning, generative modeling, sampling, test-time guidance, and scientific applications. Scientific discovery serves as a central proving ground for our work, without defining its full boundary.
Philosophy
Our research begins with intuitions and observations, uses mathematics and theory to describe and analyze them, translates the resulting understanding into applicable methods, and uses empirical evidence to refine the intuitions and theories behind those methods.
Important scientific and engineering problems live in structured spaces shaped by geometry, symmetry, hierarchy, constraints, and measurable observables. We seek to build AI methods that can use these structures to move beyond data-driven methods toward meaningful generation, exploration, and discovery.
We work in a loop connecting intuition, mathematics, computation, and evidence. Starting from structural intuitions and scientific observations, we use mathematics and theory to describe and analyze the underlying structures. We translate this understanding into implementable methods, and use empirical results to refine both the methods and the understanding behind them.
We develop foundations and algorithms for representing and generating data through Structure-Aware Learning and Modeling. We evaluate our methods through demanding problems in scientific discovery and engineering.