TC01Tutorial

This tutorial will bring together experts from various fields (ion and electron microscopies, scanning probes, atom probe tomography, chemical imaging, etc.) to discuss and present strategies for analysis of time-series, and time-frequency data using multivariate methods, Bayesian inference and wavelet transform. Relating extracted parameter fields to underlying physics will also be discussed. The tutorial will act as a forum to present achievements, share pertinent “how-to” information and discuss strategies for scalable data analysis. The aim is to identify a set of approaches that will significantly increase the quality of information extracted from imaging data, discuss their implementation and highlight the strengths and weaknesses of various methods when dealing with a particular type of problem. At the same time, identifying and sparking potential collaborations and discussions among researchers in the areas of statistics, computational sciences and materials:

8:30 am – 10:00 am
Part I: Aaron Gilad Kusne
An Introduction to Unsupervised Learning

In this tutorial we will explore a range of topics from unsupervised learning. We will begin by first discussing the differences between supervised and unsupervised learning. Next we'll briefly discuss dissimilarity measures and how to exploit them to achieve improved data analysis. We will then dive into a few data analysis techniques for visualizing high dimensional, approximating or smoothing complex data, and identifying hidden predictive latent variables. This will be followed by a look at matrix factorization techniques for identifying constituent components in data. A few clustering techniques will then be introduced and demonstrated for sorting samples into groups by similarity, identifying representative samples and segmenting data space. We will close by looking at a technique for determining the number of clusters in a data set.

10:00 am – 10:30 am  BREAK

10:30 am – 12:00 pm
Part II: Anton Ievlev
Unsupervised Learning in Spectral Materials Data

12:00 pm – 1:30 pm  BREAK

1:30 pm – 3:00 pm
Part III: Daniel Samarov
An Introduction to Supervised Learning

3:00 pm – 5:00 pm
Part IV: Stephen Jesse
Machine Learning for Image and Hyperspectral Data

Instructors

• Stephen Jesse, Oak Ridge National Laboratory
  
• Anton Ievlev, Oak Ridge National Laboratory
  
• Daniel Samarov, National Institute of Standards and Technology
  
• Aaron Gilad Kusne, National Institute of Standards and Technology