"Probabilistic modeling of stochastic plasma phenomena using information geometry," by Geert Verdoolaege (Ghent University)​

Date: 
Thursday, June 11, 2015 - 1:00pm to 2:00pm
Series: 
Plasma Seminars

Plasma Seminar

Physics & Astronomy Building (PAB) Room 4-330
Thursday, June 11, 2015
1-2PM
Refreshments at 12:30PM

Guest Speaker: Geert Verdoolaege (Ghent University)​

Talk Title: "Probabilistic modeling of stochastic plasma phenomena using information geometry"

Abstract:

The description of many complex systems in various areas of science is affected by considerable uncertainty. This can be due to incomplete physical or sensor models and measurement noise, and the uncertainty can effectively be quantified by means of probability distributions. Modern techniques from probability theory and machine learning have penetrated into the analysis of these systems, and in some cases have been used to characterize fluctuation of physical quantities. In this talk, I will discuss and demonstrate several probabilistic techniques that we have been developing and applying for the analysis of data from fusion experiments. I will start by introducing the frameworks of information geometry and Bayesian probability theory, on which our methods are based. Then I will show a new technique for estimating scaling laws in a very robust way from complex, heterogeneous data sets. Next, characterization of the stochastic properties of some classes of plasma instabilities will be discussed and estimation of plasma fluctuation properties from reflectometry will be shown. Finally, the wide applicability of the techniques will be demonstrated through an application to discrimination of textured color images.

Location: 
PAB 4-330