Skip to main content

Foundations of Machine Learning, Data Analysis, and Statistics

Exploring the principled and automated learning frontier
Adapted from C. Graziani, P. Tzeferacos, D. Q. Lamb, and C. Li, Inferring morphology and strength of magnetic fields from proton radiographs,” Review of Scientific Instruments, 88, 123507 (2017); https://​doi​.org/​1​0​.​1​0​6​3​/​1​.​5​0​13029

Argonne’s Mathematics and Computer Science Division is researching fundamental aspects of computer vision, data analysis, machine learning, imaging, statistics, and algorithmic differentiation. Our research enables the extraction of insights and construction of scientifically rigorous predictive models from computational, experimental, and observational data.

We devise techniques for automating data analysis and inference and conduct fundamental research in statistical and stochastic methods. We develop unsupervised, (semi)supervised, and reinforcement learning models and methods for regular and irregular domains incorporating domain knowledge, physical models, and constraints. We develop data science approaches and abstractions to facilitate large-scale analysis and statistical inference from rich sources of data, including multiple sensing modalities, spatiotemporal processes, and diverse phenomena. Our algorithmic research in scalable, automated machine learning streamlines method design and development using leadership-class supercomputers. We develop goal-oriented learning methods using mathematical optimization, active learning, and reinforcement learning for design of experiments in unique scientific settings such as light source facilities, the nation’s energy and electricity grid, and leadership-class supercomputers.