Artificial intelligence holds enormous promise for accelerating scientific discovery, yet its “black box” nature and data-hungry architectures often hinder adoption in domains where interpretability, physical consistency, and sparse data prevail. This course addresses these challenges by developing a unified framework that integrates three complementary, domain-informed representations—topological features, physics-based constraints, and higher-order relational structures—into end-to-end, scalable AI models. We will learn ways of integrating representations from the multiscale complexity found across scientific domains in physics, chemistry, biology, social science etc. In order to do so we will learn about methods in topological machine learning) where the data has global shape and connectivity (e.g., pore networks in materials or circulation loops in climate systems), the physics-informed AI that enforces governing laws (from fluid dynamics and stress-strain behaviors to biochemical kinetics), and the relational interactions through graph based machine learning (such as protein complexes, neural assemblies, or social group dynamics). This course will involve a research project.
Prerequisites: PSTAT120A-B, CMPSC 165A-B or equivalent. Once the quarter starts, instructor approval is required to maintain enrollment in the course, including if students do not have the listed pre-requisite courses completed.