website flavor banner

Paul Voyles

Paul VoylesPaul Voyles
Professor, Department of Materials Science and Engineering
University of Wisconsin-Madison, Wisconsin

Paul M. Voyles is a Professor in the Department of Materials Science and Engineering at the University of Wisconsin, Madison. He holds a BA from Oberlin College and a Ph.D. from the University of Illinois, Urbana-Champaign. Following a post-doc at Bell Labs, he joined UW-Madison in 2002. His research focus is the structure of materials investigated with electron microscopy, including fluctuation electron microscopy studies of the structure and dynamics of glassy materials, the study of crystal defects using high resolution STEM, and method development for quantitative electron microscopy.




Paul M. Voyles
Materials Science and Engineering, University of Wisconsin-Madison, 1509 University Ave, Madison, WI 53706

I will suggest three areas in which data scientists and materials scientists working in structural characterization might find opportunities to collaborate. Each area will be illustrated with examples from my own research or from the literature. (1) Structure optimization: For systems with a large number of structural degrees of freedom, such as nanostructures, interfaces, and non-crystalline materials, single experimental data sets rarely uniquely determine a material’s structure. Tools from statistics and optimization make it possible to combine multiple experimental data sets with information from simulations to find likely structures. (2) Discovery of new signals in large data sets: Complex, multidimensional data sets from new scanned probe, optical, x-ray, and electron microscopies and spectroscopies require new approaches to data analysis. Dictionary learning or tools for low-dimensional representation can improve analysis and lead to discovery of new signals. (3) Denoising and improved data quality: Minimally invasive characterization is required to determine the structure of radiation-sensitive materials or to sense delicate quantum states, but the price is often poor signal to noise ratio or other defects. Recent developments in compressed sensing, sparsity, and non-local means can help mitigate these limitations, making new experiments feasible.