01
Over-parameterized regime
Generalization in complex models
We study why highly parameterized models can generalize, using generalized aliasing to clarify the relationship between data, representations, and prediction.
Explore generalizationMathematics · Data · Physical systems
Data-Driven Modeling Group
What can data teach us about the models we trust?
We use data to understand, construct, and improve mathematical models—from explaining generalization in over-parameterized systems to discovering interpretable physical mechanisms.
One regimeOver-parameterizedHow does generalization emerge?
Another regimeInterpretableWhat mechanism explains the data?
Complex terrain. Useful structure.
Our work looks for the paths that connect observations to understanding.
Three active programs
Different questions.
One central idea.
Data do more than fit models. They can expose why models work, show where they fail, and help us build better ones.
02
Interpretable regime
Interpretable models from data
We combine continuous data assimilation and optimization to estimate parameters, discover equations, and identify physical mechanisms directly from observations.
Explore model discovery03
Discovering cloud physics
We apply interpretable model-discovery methods with a multi-institution team to reveal unresolved physical processes in cloud models.
Explore cloud physicsPeople
A collaborative group by design.
Member profiles connect people to research programs, publications, and the colleagues who bring each project to life.
Professor & principal investigator
Jared Whitehead
Brigham Young University
Jared Whitehead is a professor of mathematics at Brigham Young University. His group studies how data can be used to build, understand, and improve mathematical models, from highly over-parameterized learning systems to interpretable models of physical processes.
Teaching & curriculum
Applied mathematics as a complete educational experience.
Jared has played a substantial role in developing the curriculum for BYU’s Applied and Computational Mathematics Emphasis (ACME).
Previous work
Historical earthquake and tsunami inference
Research using historical Indonesian tsunami accounts, Bayesian inference, MCMC, and GeoClaw simulations to reconstruct source earthquakes and improve modern seismic-risk assessment.