Applied and Computational Mathematics Seminar
Machine learning of Symmetric Crystallographic Data With Separable Functions, by Ryan Botts (Ohio University)
| What |
|
|---|---|
| When |
Feb 12, 2008 from 04:10 pm to 05:00 pm |
| Where | 322 Morton Hall |
| Contact Name | Martin Mohlenkamp |
| Contact Email | mjm@math.ohiou.edu |
| Contact Phone | 740-593-1259 |
| Add event to calendar |
vCal
iCal
|
Title: Machine learning of Symmetric Crystallographic Data With Separable Functions
Speaker: Ryan Botts (Ohio University)
Abstract: One would like to be able to predict properties of crystal structures from the knowledge of the properties of structures on the same lattice, but with different arrangements of the elements. This would allow you to predict those structures that optimize desirable properties. We can pose this as a problem in multivariate regression in high dimension, with a set of symmetry and consistency constraints imposed by the lattice. Physicists have developed methods for predicting these properties based on the use of cluster expansions. The main difficulty in this approach is identifying which figures to include. Our method uses "pseudo-figures" that have no physical meaning, yet provide more accurate approximations and predictions. We will discuss a regression model that uses sums of separable functions and an Alternating Least Squares approach.
