Low-dimensional Representations of High-dimensional Labelled Glass Data

Dr. Nicholas Mecholsky
Research Scientist, Vitreous State Laboratory
 Research Associate Professor, Physics Department
The Catholic University of America, Physics Department

Wed, April 16, 2025 - 4:00 PM
In Hannan 108 (Auditorium)

gk-sm.jpgFinding low-dimensional representations of natively high-dimensional data is a central problem in science. In our work at the Vitreous State Laboratory, we often encounter high-dimensional data in the form of oxide compositions of glasses based off of complicated nuclear waste mixtures, for example at the Hanford Site Tank Farms. In addition to high-dimensional data, separating data by using labels (such as whether a particular crystal forms when cooling the glass) adds more complexity to an already difficult problem. In this work I will explore the concept of subspaces of high-dimensional spaces and introduce the notion of the Grassmannian. I will then use this to describe a method for finding the best subspace in which to separate labeled data of glasses.

           Low-dimensional representations of high-dimensional labelled glass data

If you have any questions about the Colloquium Series, would like to request disability accommodations  or would like to make a donation please contact the Physics Department, cua-physics@cua.edu or  call (202) 319-5315.