Aug 4, 2013 — Dimensionality Estimation of ICA with Bayesian Analysis — Across methods of matrix decomposition (ICA, Archetypal Analysis, NNMF), we run into the problem of needing to know how many signals to decompose our data to.  This is called model order selection.  If we choose a number that is too... Read more

Aug 3, 2013 — Mutual Information for Image Registration — Mutual information is an important concept for medical imaging, because it is so prominently used in image registration.  Let’s set up a typical registration problem.  Here we have two pictures of something that you see everyday, a dancing pickle! I’ve... Read more

Aug 3, 2013 — Multidimensional Scaling (MDS) — Multidimensional scaling (MDS) is a way to reduce the dimensionality of data to visualize it.  We basically want to project our (likely highly dimensional) data into a lower dimensional space and preserve the distances between points.  The first time that... Read more

Jul 31, 2013 — Imputation of Missing Values — In a perfect world, my data is absolutely perfect, and I’ve collected every feature or metric for every person, to be fed cleanly into some model.  Now let’s recognize the obvious reality that having complete data is rarely the case.... Read more