Yao Zhang - Scientific Machine Learning, SciML

Scientific machine learning (SciML) integrates data-driven approaches with scientific knowledge to build predictive, reliable, and interpretable models. It begins with core data science practices, including data acquisition and cleaning, feature and representation learning, uncertainty quantification, and rigorous model evaluation to ensure robustness beyond curated datasets. Building on this foundation, physics-informed neural networks embed governing equations, boundary conditions, and conservation laws into training so that models respect known physical principles. Neural operator methods extend this idea by learning mappings between function spaces, providing efficient surrogates for parametric partial differential equations. In parallel, causality-aware machine learning introduces causal inference to distinguish correlation from intervention effects, improving generalization under confounding and distribution shifts. Advances in deep learning optimization and bio-inspired algorithms further enable scalable training, global search, and effective model design for complex scientific applications.